[{"data":1,"prerenderedAt":3898},["ShallowReactive",2],{"changelog-static-v1":3},{"schema_version":4,"release_versions":5,"entries":25},1,[6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24],"12.1.34","12.1.33","12.1.32","12.1.31","12.1.30","12.1.29","12.1.28","12.1.27","12.1.26","12.1.25","12.1.24","12.1.23","12.1.22","12.1.21","12.1.20","12.1.18","12.1.16","12.1.15","Unreported",[26,36,46,54,62,70,77,85,94,105,116,126,134,143,153,161,169,177,186,193,201,207,215,222,229,241,249,258,270,278,289,297,305,315,324,334,342,353,360,367,374,381,389,396,403,410,417,424,431,438,445,452,459,466,474,481,488,497,505,513,521,529,537,544,551,558,565,573,580,588,595,602,609,616,623,630,638,647,656,663,673,680,688,696,704,713,720,727,734,740,748,757,765,772,779,787,795,803,811,819,827,835,841,847,853,859,869,877,885,891,897,903,911,924,932,940,951,959,968,975,984,993,1001,1008,1016,1025,1031,1039,1046,1052,1059,1065,1071,1077,1083,1089,1095,1108,1115,1121,1128,1135,1142,1149,1156,1162,1169,1175,1182,1188,1194,1201,1210,1218,1225,1232,1239,1245,1251,1257,1263,1269,1275,1281,1289,1295,1301,1307,1319,1326,1332,1338,1347,1353,1360,1366,1375,1382,1388,1396,1403,1409,1415,1422,1429,1435,1441,1448,1455,1462,1469,1477,1483,1490,1496,1504,1512,1520,1527,1535,1542,1549,1559,1567,1576,1585,1594,1601,1608,1615,1622,1630,1639,1646,1652,1659,1666,1674,1683,1689,1700,1707,1714,1722,1730,1738,1746,1756,1763,1772,1779,1787,1792,1800,1808,1815,1822,1829,1837,1844,1850,1856,1862,1869,1878,1887,1895,1904,1912,1919,1927,1935,1943,1950,1958,1965,1971,1979,1986,1993,2001,2009,2017,2025,2032,2039,2046,2053,2061,2069,2078,2088,2096,2103,2111,2121,2131,2140,2149,2156,2163,2171,2186,2194,2202,2211,2230,2248,2257,2265,2274,2282,2290,2298,2304,2312,2320,2326,2332,2339,2345,2352,2362,2371,2379,2389,2398,2407,2416,2425,2435,2445,2454,2468,2477,2487,2497,2505,2513,2522,2531,2540,2561,2571,2580,2588,2597,2612,2621,2631,2642,2653,2662,2673,2684,2693,2704,2712,2723,2735,2744,2752,2760,2768,2776,2785,2794,2806,2817,2827,2853,2868,2881,2895,2910,2933,2954,2965,2976,2986,2997,3008,3018,3028,3040,3048,3056,3065,3075,3097,3112,3121,3143,3157,3196,3209,3218,3230,3237,3246,3255,3265,3281,3294,3303,3312,3320,3331,3355,3365,3380,3395,3414,3426,3435,3446,3464,3476,3489,3502,3515,3523,3535,3548,3555,3562,3574,3585,3594,3604,3614,3626,3631,3645,3654,3662,3669,3681,3692,3706,3725,3731,3744,3751,3759,3765,3774,3787,3799,3808,3828,3838,3849,3856,3863,3870,3877,3888],{"version":27,"date":28,"sections":29},"20.2.13","2026-10-23",[30],{"name":31,"changes":32},"Changed",[33,34,35],"Recognize the verified Crystal story yes\u002Fno call chain as dialogue while preserving conservative native UI ownership.","Allow one freshly qualified Red final-heal trip after policy training, within the original preparation recovery limits.","Review two exact historical maintenance contention failures without clearing unrelated failures, and defer only mutation-free admission contention.",{"version":37,"date":38,"sections":39},"20.2.12","2026-10-22",[40],{"name":31,"changes":41},[42,43,44,45],"Start Red Pewter Gym recovery at the settled exit tile, preventing repeated Gym re-entry during healing and PC-deposit trips.","Honor declared Crystal interior exits during bounded stall rescue while preserving native UI ownership and rescue budgets.","Migrate only the exact admitted Red Gym recovery branch through managed publication; preserve learner weights, Adam state, rewards, rules and qualification gates.","Review recovered, unpublished preflight capacity retries against a later complete operator-pause checkpoint, preserving failed evidence and all other failure guards.",{"version":47,"date":38,"sections":48},"20.2.11",[49],{"name":31,"changes":50},[51,52,53],"Add an exact reviewed resume for completed operator pauses after historical optimizer closeout, preserving all ordinary launch and integrity guards.","Bind resume reviews to current control source, campaign ownership, saved learner state and the published release candidate.","Retry initial read-only publication resource binding within a bounded window when a scheduled coordinator owns the campaign lock.",{"version":55,"date":38,"sections":56},"20.2.10",[57],{"name":31,"changes":58},[59,60,61],"Upgrade the web runtime and dependency tree to patched Node 24, Nuxt, Vue, Router, Wrangler and compatible tooling releases.","Upgrade PyTorch to 2.14.1 and Gymnasium to 1.4 with exact macOS Apple Silicon Python 3.11 dependency sets and updated security dependencies. Pin PyBoy 2.7.0 after 2.8.1 fails retained Red\u002FCrystal save-state loading.","Document separate dependency validation, installation and managed campaign activation while preserving existing learner and optimizer lineage.",{"version":63,"date":38,"sections":64},"20.2.9",[65],{"name":31,"changes":66},[67,68,69],"Share campaign review-file I\u002FO and prepared publication mechanics while preserving exact per-command and recovery safeguards.","Move reviewed configuration migrations into a dedicated module and bind extracted release helpers to native source inventories.","Avoid repeated YAML parsing during semantic verification with a bounded exact-content cache, fresh reads and independent returned mappings.",{"version":71,"date":38,"sections":72},"20.2.8",[73],{"name":31,"changes":74},[75,76],"Centralize source and installed-distribution release checks and add a previewable command for synchronized numeric release preparation.","Validate publication configuration options consistently so receipt recovery and worker expansion cannot silently ignore reviewed behavior changes.",{"version":78,"date":38,"sections":79},"20.2.7",[80],{"name":31,"changes":81},[82,83,84],"Defer typed campaign and resource lock-acquisition contention without terminating a live optimizer comparison; preserve holds for integrity faults and ambiguous operation failures.","Add exact managed publication review and incomplete closeout for the interrupted Crystal comparison. Preserve its failed evidence, reserved storage, charged work and ordinary learner; resume only through the reviewed transaction without retrying measured arms.","Distinguish incomplete comparison closeout from successful completion and keep ordinary learning, source activation and protected checkpoint qualification separate.",{"version":86,"date":38,"sections":87},"20.2.6",[88],{"name":31,"changes":89},[90,91,92,93],"Copy frozen native campaign authority metadata through the existing primitive serializer when preparing optimizer trial arms, preserving storage\u002Fpolicy contracts and parent state.","Cover both frozen storage-binding and policy-contract boundaries with real-Authority regressions. Preserve the failed unmeasured20.2.5 preparation and its durable hold; this source fix grants no retry or activation authority.","Add the approved, exact reviewed recovery for the unmeasured pre-authority import failure. Separate source publication authorization from candidate-bound disposition, retaining the failed attempt, original reservation and distinct ordinary\u002Fexperiment bundles.","Permit one matched successor only after the declared managed source handoff completes. Preserve starter minimum, semantic and capacity gates; reject measured retries and ordinary-training fall-through, and retire only the completed recovery's own restrictions.",{"version":95,"date":38,"sections":96},"20.2.5",[97],{"name":31,"changes":98},[99,100,101,102,103,104],"Budget poisoned recovery through the complete nurse trip, retain emergency native Potion receipts, and add bounded medicine replenishment with exact cash\u002Finventory proof.","Retire authenticated stale Red recovery receipts after settled authored exits while preserving learned graphs, failure counts and finite route budgets.","Check team PP allocation, incoming trainer damage and actual XP recipients; preserve a viable Crystal replacement before prospective Gym quarantine.","Add narrowly authenticated Crystal Route 30 trainer execution, saved quest return after recovery\u002Fshop detours and an explicit exhausted-rescue disposition.","Add managed sequential matched Crystal optimizer comparisons with preserved policy\u002FAdam\u002Fmemory inputs, bounded write credits and separate ordinary\u002Fprotected checkpoint authority. Keep ordinary PPO defaults unchanged.","Gate newly assisted routes on native evidence and record source, installation, publication, activation and experiment results separately in the release notes.",{"version":106,"date":107,"sections":108},"20.2.4","2026-10-21",[109],{"name":31,"changes":110},[111,112,113,114,115],"Add bounded native Crystal Cherrygrove Ball restocking with paired cash\u002Finventory receipts, explicit controller ownership and durable episode evidence. Keep Elm gifts and unverified inventory changes separate from purchases.","Add reviewed Red preparation retreat and surveyed recovery priority without waiving battle forecasts, level caps, spent attempts or native transition settling. Extend Crystal antidote and audited Center recovery handling.","Reconcile policy\u002Fcontroller\u002Funknown ownership, separate unique rollout transitions from attempted\u002Fapplied PPO exposures, and retain native party-PP and delayed move-intent evidence without inventing missing results.","Add named later-milestone diagnostics with a 500k horizon, explicit missing\u002Fcensored\u002Fassisted coverage and immutable development\u002Fheld-out comparisons. Prepare separate matched Crystal ten-to-one epoch experiment recipes without changing campaign optimizer defaults or granting a second trainer.","Restrict default pytest discovery to maintained unit tests and expose native, retained and qualification tiers. Record source, installation, publication, native verification and optimizer activation separately in the release notes.",{"version":117,"date":118,"sections":119},"20.2.3","2026-10-20",[120],{"name":121,"changes":122},"Changes",[123,124,125],"Admit the measured complete historical launcher inventory with a finite 768 MiB aggregate scan-work allowance. Preserve the 16 MiB per-file memory bound, 1,024-entry bound, cancellation deadlines and every custody\u002Fhash check.","Check the full consumer-control history before reserving or copying publication bytes, then retain the existing fresh verification at native sealing. Preserve the complete failed 20.2.2 capsules and their accounting.","Carry forward the reviewed Red settled preparation-phase activation and bounded navigation diagnostics. Preserve learner, Adam, memory, rewards, PPO, source-coverage gates and protected-checkpoint qualification.",{"version":127,"date":118,"sections":128},"20.2.2",[129],{"name":121,"changes":130},[131,132,133],"Prepare the existing settled Red preparation-phase guard through an explicit reviewed publication option. A transient south-gate map\u002Fcoordinate pair must not permanently select Gym preparation; preserve readiness floors, spent budgets, Forest policy handoff and Brock entry checks.","Retain Red Route 1 rescue outcomes and bounded learned-planner failure counts in failure capsules, compact contexts and terminal CSV\u002FJSON diagnostics. Preserve the difference between a spent intervention, a blocked route and absent assistance.","Preserve rewards, PPO settings, learned parameters, Adam state, auxiliary memory and lineage. Record source, editable installation, candidate publication and managed optimizer activation separately.",{"version":135,"date":118,"sections":136},"20.2.1",[137],{"name":121,"changes":138},[139,140,141,142],"Yield native Crystal travel dialogue before nurse-route inputs, field items or readiness-trip charges. Resume the same bounded recovery after policy closes the text; preserve nurse transaction ownership, unknown-UI timeouts and spent attempts.","Share the existing Elm-return dialogue handoff with ordinary Cherrygrove recovery, covering the recorded post-gift Route 29 input-owner timeout.","Admit source packaging from fully saved managed pauses through snapshot custody, holding trainer exclusions and rechecking the original pause at both storage write boundaries. Preserve rollout holds, optimizer integrity, capacity accounting and interrupted-stage evidence.","Preserve reward and PPO settings, team-preparation safety stops, learner weights, Adam state, auxiliary memory and lineage. Record source, installation, publication and managed activation separately.",{"version":144,"date":145,"sections":146},"20.2.0","2026-10-19",[147],{"name":121,"changes":148},[149,150,151,152],"Recognize native Red trainer-turn progress during deferred finisher synchronization, so a resolved KO and next-opponent transition do not become a false safety-switch failure. Preserve battle identity, fresh legality checks and finite waits.","Accept both existing ROM-verified Crystal overworld phases during nurse navigation and lead rotation. Keep unknown UI, script\u002Fentry ownership and movement guards intact.","Add opt-in bounded Crystal menu cancellation on exact audited quest-return routes. Verify Cherrygrove's native return receipt before assisting its shared Center 2F; preserve spent rescue and menu budgets.","Preserve rewards, PPO settings, learner weights, Adam state, auxiliary memory and model lineage. Record source, editable installation, candidate publication and managed optimizer activation separately.",{"version":154,"date":145,"sections":155},"20.1.15",[156],{"name":121,"changes":157},[158,159,160],"Prepare opt-in Red ownership of the same unsafe wild encounter through native escape after preparation exhaustion. Preserve failed sessions, finite budgets, actual XP bounds and Gym entry protections.","Prepare opt-in Crystal post-gift readiness recovery with bounded nurse trips, authenticated lead rotation and the surveyed Cherrygrove house exit. Preserve retired identities, native UI ownership and quest-rescue limits.","Keep rewards, PPO settings, learned weights, Adam state and model lineage unchanged. Record source, installation, diagnostics, publication and live activation separately.",{"version":162,"date":145,"sections":163},"20.1.14",[164],{"name":121,"changes":165},[166,167,168],"Prepare bounded retries for the exact pre-work native resource-admission lock conflict during managed learner startup. Preserve nonblocking role guards, cooperative deadlines, integrity failures and normal recovery limits.","Retain the 20.1.13 Crystal gate travel fix and Red guarded Junior result-settling configuration. Keep learned weights, Adam state, memory, rewards and source-coverage requirements unchanged.","Keep source, installation, candidate publication and optimizer activation separate. The existing Crystal startup hold is preserved pending a supported reviewed recovery.",{"version":170,"date":145,"sections":171},"20.1.13",[172],{"name":121,"changes":173},[174,175,176],"Preserve Crystal's reserved-XP PC travel across the surveyed Route 31\u002FViolet gate so Gym preparation cannot send the same party back through the gate. Keep native evidence, actual trainer XP, quarantine and finite route checks.","Prepare the existing guarded Red Junior Trainer result-settling mode for managed activation. Wait for the ROM's defeat flag before recomputing its spent XP reserve; preserve cap, party, caller-proof and wait limits.","Preserve rewards, PPO settings, learned weights, Adam state and checkpoint lineage. Source, installation, publication and live optimizer activation are recorded separately in the release evidence.",{"version":178,"date":179,"sections":180},"20.1.12","2026-10-18",[181],{"name":121,"changes":182},[183,184,185],"Keep completed Red Forest and clinic policy handoffs active through optional field interiors; recheck the full Brock forecast at the actual Gym boundary without renewing spent preparation budgets.","Add opt-in Crystal field headroom handling so a currently legal attacker can seek teammates instead of stopping solely on all future Falkner-trainer XP reservations. Preserve actual over-cap, permadeath, ordinary-trainer XP and Gym-entry guards.","Count a ROM-proven phone input advancing to a new native script instruction as progress during Crystal PC travel, while preserving frozen-wait and total route limits. Expose the field flag to bounded frozen diagnostics without changing primary rewards or promotion authority.",{"version":187,"date":179,"sections":188},"20.1.11",[189],{"name":121,"changes":190},[191,192],"Advance only native ROM-verified Crystal paragraph\u002Fscript input waits during PC travel, including the exact first Route 31 Mom saving-money offer. Preserve finite route budgets, quarantine, cap checks and controller credit.","Prepare the existing qualified Red clinic-travel option so a missing Brock fight forecast does not block native healing travel. Preserve the failed session, learner, Adam state and protected checkpoint.",{"version":194,"date":195,"sections":196},"20.1.3","2026-10-16",[197],{"name":121,"changes":198},[199,200],"Add independently controlled Red fixes for settled preparation phases, clinic travel without a full Brock PP forecast, native recovery UI\u002Fblocked-edge evidence, and Junior Trainer result settlement. Preserve cap, finite budgets, actor masks and optimizer lineage; new switches remain disabled until managed diagnostic review.","Qualify a bounded native Viridian School exit and retain diagnostic evidence for Forest stalls and premature Gym cap decisions. Existing Forest trainer-readiness activation remains unsupported by its observed trigger and weak-party coverage.",{"version":202,"date":195,"sections":203},"20.1.2",[204],{"name":121,"changes":205},[206],"Count completed, uncensored recovery-navigation failures toward per-starter source-release coverage, preserving their original invalid gameplay outcomes and all source, starter, legality and integrity checks.",{"version":208,"date":209,"sections":210},"20.1.0","2026-10-15",[211],{"name":121,"changes":212},[213,214],"Prepare Crystal's post-gift Route 29\u002F30 journey with native legal capture inputs, bounded safe wild training and surveyed nurse recovery, preserving the eight-action policy and native Nuzlocke guards.","Count Crystal Ball inventory consumption with Crystal item IDs and distinguish it from policy macro requests. Preserve learned weights, optimizer lineage and managed release activation requirements.",{"version":216,"date":209,"sections":217},"20.0.63",[218],{"name":121,"changes":219},[220,221],"Separate exhausted Red Forest travel capability from the conservative Brock\u002FJr. Trainer PP forecast. A living damaging attacker can restore health\u002Fstatus\u002FPP and traverse the surveyed southern corridor with the existing four-PP floor, while northern Gym travel retains its full fight forecast.","Record the fallback PP scope in diagnostics. Preserve native trainer-entry checks, death and encounter history, level caps, actor masks, finite recovery budgets and the failed preparation session.",{"version":223,"date":209,"sections":224},"20.0.62",[225],{"name":121,"changes":226},[227,228],"Wait for brief campaign transaction locks during read-only runtime capacity-fence inventory using existing native bounded acquisition and the original proof deadline. Preserve changed-dependency holds and one-shot resource writes.","Carry forward exact retained-stage recovery, Red Forest handoff, and the unchanged paired Crystal quest configuration.",{"version":230,"date":209,"sections":231},"20.0.61",[232],{"name":121,"changes":233},[234,235,236,237,238,239,240],"Recover an exact complete retained ordinary runtime staging peak through native full-payload verification and capacity custody, preserving its files and source identity without publication or learner rights.","Retry typed read-only capsule-completion contention before the resource write; never replay an attempted write or weaken integrity failures.","Include the verified Red Forest handoff fix and preserve the exact paired Crystal quest-followthrough configuration.","Batch navigation peer safety and edge probes, reducing each expanded-position peer read from twelve statements to two while preserving merge order, bounded outcomes and fresh read transactions.","Invalidate only local navigation partitions at the changed source position, including every context there; preserve unrelated cached partitions.","Add bounded rollout\u002FPPO\u002Fworker timing diagnostics with passive frame counts and active source identity. Diagnostics remain opt-in pending native overhead qualification.","Add isolated navigation and recurrent-update replay tools. Crystal thread screening did not justify a runtime or PPO change. Component benchmarks do not establish live episode throughput improvement.",{"version":242,"date":209,"sections":243},"20.0.60",[244],{"name":121,"changes":245},[246,247,248],"Complete Red exhausted-recovery inputs through the surveyed Forest gate and entrance apron, using native UI and transition readiness.","Return Forest traversal to policy after an observed owned gate entry; preserve spent recovery budgets and resume the failed preparation handoff at the northern route without resetting its session.","Preserve the prepared Crystal quest-followthrough configuration, learned parameters, Adam state, auxiliary memory and existing measured evaluation jobs through the managed paired release.",{"version":250,"date":209,"sections":251},"20.0.59",[252],{"name":121,"changes":253},[254,255,256,257],"Preserve finite Crystal return-phase budgets after the verified one-use opening rival loss and native healing, without recording a victory or weakening permadeath elsewhere.","Extend audited post-gift policy guidance and bounded rescue across New Bark detours, Route 29 and Cherrygrove. Keep exact native Egg\u002FBall receipts, map-local fields, unknown UI exclusions, controller actor masks and monotonic attempt limits.","Measure Crystal quest comparisons through the Ball gift and ordered native connections back to Route 30, with version 2 native UI phase and action-owner counts. Ordinary opening qualification retains its existing scope.","Preserve Red's resolved gameplay settings, learned parameters, Adam state, auxiliary memory and historical holds through the managed paired release channel.",{"version":259,"date":260,"sections":261},"20.0.58","2026-10-14",[262],{"name":121,"changes":263},[264,265,266,267,268,269],"Reserve damaging PP for Red's undefeated Pewter Junior Trainer as well as Brock, retaining complete native XP and level-cap obligations.","Add a finite, default-off recovery and Gym handoff for an exhausted preparation session with a verified usable lead. Preserve strict whole-party readiness and stop with a named inability when no legal route attacker remains.","Add default-off Red signed-navigation credit deduplication across uncharged recovery retreats, retaining reward magnitudes, ordinary signed movement, and Crystal behavior.","Extend only Red frozen behavior diagnostics through the first badge, with native Forest\u002FPewter\u002FGym exposure, Brock result and badge-transition evidence. Keep ordinary opening qualification and protected promotion separate.","Preserve verification of authenticated historical failed Red panels when their frozen configs omit the later default-false Crystal evaluation metadata.","Preserve compatible learned parameters, Adam state, auxiliary memory and existing incidents through the managed source-release channel.",{"version":271,"date":260,"sections":272},"20.0.57",[273],{"name":121,"changes":274},[275,276,277],"Allow one Crystal starter faint in the native pre-Poké Ball Cherrygrove rival fight, then let the original defeat script heal the party and continue.","Share the one-use allowance between ROM faint\u002Fwhiteout guards and reward tracking. Repeated hook callbacks cannot spend another allowance; existing deaths, wild battles, later rivals and depleted Ball inventory retain ordinary permadeath.","Keep primary evaluations on the resolved training rules and disable the Crystal exception in the explicitly named strict challenge profile. Preserve compatible policy weights, Adam state and auxiliary memory through the managed campaign release channel.",{"version":279,"date":280,"sections":281},"20.0.56","2026-10-13",[282],{"name":121,"changes":283},[284,285,286,287,288],"Add default-off native Crystal rival execution and finite pre-rival readiness, separate from Falkner preparation; retain the level-7\u002Ffull-HP\u002F10-damaging-PP candidate and strict XP headroom checks.","Add default-off Forest entry readiness and finite Pewter recovery planning under the existing native gym cap contract.","Add opt-in Crystal event-bound rescue budgets and surveyed post-gift Route 29\u002FCherrygrove scopes, preserving legacy behavior when disabled.","Add create-only frozen behavior diagnostic pairs, separate from ordinary evaluation\u002Fpromotion, with exact bundle\u002Fruntime\u002Fasset bindings, singleton host admission and no measured retry.","Add a disposable native fixture producer. Gameplay features remain disabled pending native evidence and first-attempt comparison review; this note records source implementation, not activation.",{"version":290,"date":280,"sections":291},"20.0.55",[292],{"name":121,"changes":293},[294,295,296],"Publish bounded current worker episode snapshots with exact episode identity, native steps, starter, location, milestones, badges and party state before an episode finishes.","Add independently refreshed Current episodes panels to Red and Crystal Atlas and Analysis, preserving completed-run counts and explicitly reporting stale, paused and unavailable evidence.","Bind campaign observations to their selected execution, source release and model, reject older observations, and retain existing heartbeat upload limits and managed learner\u002FAdam handoffs.",{"version":298,"date":280,"sections":299},"20.0.54",[300],{"name":121,"changes":301},[302,303,304],"Defer managed campaign cron ticks only when native admission lock acquisition is busy, retaining the existing learner and next scheduled tick instead of reporting an exception.","Preserve integrity errors, unknown I\u002FO faults, generic transaction failures after writes, native lock cleanup, and unchanged Red\u002FCrystal learning configurations.","Add real-lock regression coverage for selector, coordinator, maintenance, status and notice preflights, including successful subsequent dispatch and cron exit status.",{"version":306,"date":280,"sections":307},"20.0.53",[308],{"name":121,"changes":309},[310,311,312,313,314],"Defer typed native campaign-lock contention before write-credit debit and optional maintenance dispatch, preserving the learner, quota credits and existing integrity\u002Fwrite failure handling.","Add exact reviewed recovery for the approved Red source 52 write-budget contention incident, preserving its failure evidence and accounting for 144,384 unsaved optimizer steps.","Read native launcher registrations and paired quarantine review outputs using their existing 16 MiB envelopes; retain unrelated metadata and aggregate inventory limits.","Recognize the exact native mount lock during launcher discovery and preserve byte-level references in historical binary controls.","Keep reviewed quarantine reversible and fully charged, with unchanged source 20.0.52 learning configurations and managed learner\u002FAdam handoffs.",{"version":316,"date":280,"sections":317},"20.0.52",[318],{"name":121,"changes":319},[320,321,322,323],"Add opt-in, bounded navigation imitation from verified successful waypoint steps in each game's current rollout. Retain sampled PPO actions, controller ownership masks, recurrent context, model shapes and optimizer lineage.","Reject unsafe or incomplete examples, apply supervision only once per rollout, and record accepted examples separately from supervision actually applied after the KL guard. Exclude teacher and padded rows before PPO ratio exponentiation.","Bind the resolved objective on fresh, resumed and coupled campaign learners; preserve default-off behavior and assisted-episode qualification exclusions.","Include the unpublished 20.0.50 within-game memory improvements and 20.0.51 managed publication and retention changes in the coordinated release.",{"version":325,"date":280,"sections":326},"20.0.51",[327],{"name":121,"changes":328},[329,330,331,332,333],"Save paired publication preparations before selecting candidates, and resume exact unfinished publication or finish existing candidates without rebuilding unchanged sealed runtimes.","Add reviewed reversible quarantine for unused publication runtimes and launchers, retaining provenance and charged storage while protecting active, pending and historical consumers.","Reconcile exact legacy duplicate runtime reservations with immutable proof, preserving payload bytes and ordinary overlap rejection.","Support native-drive SQLite archive controls and correctly project the admitted large-drive limits into maintenance status. Enable bounded retention through existing exact policy reviews.","Carry the independently verified within-game memory and waypoint-assistance changes from the unpublished 20.0.50 source into the coordinated release, preserving qualification exclusions.",{"version":335,"date":280,"sections":336},"20.0.50",[337],{"name":121,"changes":338},[339,340,341],"Validate active trainer slots before recording opponents and keep uncertain historical observations out of team-size and PP calculations. Preserve raw evidence and game, ROM, starter and progression boundaries.","Continue bounded, idempotent trainer species, level and confirmed-move learning after encounter accounting saturates. Preserve encounter history, counts, saturation and frozen evaluation semantics.","Add opt-in bounded waypoint assistance using observed same-map navigation, with existing context\u002Fhazard checks, controller attribution and assisted-episode qualification exclusions.",{"version":343,"date":280,"sections":344},"20.0.49",[345],{"name":121,"changes":346},[347,348,349,350,351,352],"Reduce repeated runtime-inventory filesystem calls in active-capsule attestation and the sealed bootstrap. Preserve exact member inventories, fresh file identities, symlink rejection, complete initial hashing and managed source handoffs.","Coalesce repeated native dependency verification during read-only runtime-staging inventories to one-second intervals, with unconditional checks at inventory boundaries and immediately before admission. Keep fresh usage, byte limits, custody, source checks and physical floors.","Let ordinary source publication wait briefly and rebuild its pre-staging proofs after typed native lock contention. Bound retries and preserve all integrity failures, special recovery paths, and any attempted staging write or partial preparation.","Resolve release-candidate and prepared-release storage through the validated saved-transfer binding on Extreme. Add explicit publication-time binding of the existing scheduler jobs, after all requested candidates publish, with activation left to their normal schedule and managed handoff.","Add an explicit, reasoned publication option to waive the six-completed-runs-per-starter waiting gate for the exact candidate. Record the existing source-bound reviewed override; retain every other hold, compatibility check, capacity requirement and checkpoint handoff.","Retain the admitted Extreme limits of 256 GiB per game, a 100 GiB free-space floor and a 1 GiB checkpoint margin. Current campaigns do not use the older operations-policy quota-refresh path; changing its cadence would not improve their throughput.",{"version":354,"date":280,"sections":355},"20.0.48",[356],{"name":121,"changes":357},[358,359],"Prepare a compatible controller on the native snapshot lane only for the exact human-approved, completed Red reconciliation while its learner remains paused. Verify native receipt, owner, source, latest complete bundle and immutable authorization before and after preparation.","Preserve ordinary rollout holds, original paired saved-transfer behavior, full byte\u002Fpeak accounting and physical floors. No early resume, direct trainer or Crystal restart.",{"version":361,"date":280,"sections":362},"20.0.47",[363],{"name":121,"changes":364},[365,366],"Read the exact existing launcher registration using the native 16MiB control-document bound rather than the generic 1MiB metadata bound. Keep read-only FD custody, hashes, case bindings and all other limits unchanged.","Preserve the actual source46 registration-read denial and the still-unstarted approved recovery request; no learner\u002Fhold transition or completed-control rebuild.",{"version":368,"date":280,"sections":369},"20.0.46",[370],{"name":121,"changes":371},[372,373],"Preserve and recognize only the exact native zero-valued daily operations summary that followed the approved Red failure, allowing its unchanged case to reach native review. All later work, other summaries, changed counters and integrity failures remain held.","Preserve the source45 dry-review denial; no review or recovery transaction had started, and no learner state or incident was changed.",{"version":375,"date":280,"sections":376},"20.0.45",[377],{"name":121,"changes":378},[379,380],"Add only the explicitly human-approved native reviewed Red evaluation-contention recovery, preserving the exact latest complete imported learner and all original incidents while accounting for 27,648 unsaved trained steps plus 3,072 untrained exposure.","Verify exact native interrupted transaction prefixes and current owner\u002Faccounting before continuing; no generic integrity-hold reset, direct trainer, forced job run or Crystal restart.",{"version":382,"date":383,"sections":384},"20.0.44","2026-10-12",[385],{"name":121,"changes":386},[387,388],"Defer optional evaluation housekeeping only on typed native StoreBusy contention instead of declaring learner state unsafe. Preserve genuine source, custody and unknown filesystem-error holds.","Preserve the first Extreme Red failed execution, its unsaved optimizer exposure and all imported learner\u002FAdam\u002Fmemory. This fix does not reconcile or resume any existing integrity hold.",{"version":390,"date":383,"sections":391},"20.0.43",[392],{"name":121,"changes":393},[394,395],"Charge every plain unsealed runtime_capsule reservation in full during staging capacity admission, taking no used-space credit and avoiding runtime-tree scans. Preserve separate native hashed obligations, complete campaign accounting, verifier callbacks and deadlines.","Preserve both300second preflight failures and all already imported source39 learner\u002FAdam\u002Fmemory; no deduplication, deletion, extra limits or direct launch.",{"version":397,"date":383,"sections":398},"20.0.42",[399],{"name":121,"changes":400},[401,402],"Reuse a verified immutable authority only while its native ledger signature\u002Fadmitted-event anchor and exact payload remain unchanged, avoiding repeated construction during inventory callbacks. Preserve every native verifier, deadline, capacity obligation and mutation rejection.","Preserve the actual source41 preflight timeout and all source39 saved state; no completed runtime rebuild, learner reset, direct launch or forced scheduler run.",{"version":404,"date":383,"sections":405},"20.0.41",[406],{"name":121,"changes":407},[408,409],"Classify only the fully identified unmounted EFI container partition separately from APFS training volumes in native image inventory; preserve unknown, mounted, duplicate and foreign-volume rejection.","Resolve compatible controller scope from the exact saved-transfer external mount when its immutable binding has no legacy storage_root field. Preserve source39 learner\u002FAdam\u002Fmemory\u002Fruntime and prior failed launchers.",{"version":411,"date":383,"sections":412},"20.0.40",[413],{"name":121,"changes":414},[415,416],"Project the exact saved-transfer image field into the sealed bootstrap registration as image_path, preserving the admitted authority and immutable learner runtime. Reject conflicting paths; legacy registrations remain unchanged.","Preserve the actual source39 scheduled startup failures. Repair only managed registration\u002Fjob binding; no data reimport, runtime rebuild, direct trainer or forced run.",{"version":418,"date":383,"sections":419},"20.0.39",[420],{"name":121,"changes":421},[422,423],"Read exact saved navigation\u002Ftrainer memory from native verified bundle layouts during the explicit current-state transfer; retain legacy live-path checks.","Prepare the new paired runtime through bounded snapshot admission while the original complete human-stopped learners remain excluded. No rollout or launch authority is granted by preparation.",{"version":425,"date":383,"sections":426},"20.0.38",[427],{"name":121,"changes":428},[429,430],"Support the human-directed smaller transfer of the latest complete saved learner, Adam and configured memory into a fresh Extreme managed campaign, retaining the original history\u002Fimage and all failed preparation.","Bind only the existing scheduler jobs and await their normal scheduled run through an explicit managed option; no forced run or new job. Source, installation, bundle admission and actual optimizer activation remain separate.",{"version":432,"date":383,"sections":433},"20.0.37",[434],{"name":121,"changes":435},[436,437],"Scan captured control-file reference strings as exact bytes, including binary files, instead of assuming every immutable file is UTF-8 text. Preserve full captured hashes\u002Fidentities and reject any helper path\u002Fhash reference in either representation. The actual source36 preclaim failure is retained.","Keep the same approved unused-helper retirement\u002Fone corrected successor scope, original budgets and native protections; no helper or learner data was retired by the failed preflight.",{"version":439,"date":383,"sections":440},"20.0.36",[441],{"name":121,"changes":442},[443,444],"Allow the existing native full-claim acquisition to wait up to180seconds for already-running managed helpers to drain, while holding its existing mount exclusions. Partial claims unwind on typed contention, the same outer deadline remains, and the protected retirement\u002Fbuild body is never retried.","Keep the original approved two-unused-helper\u002Fone-corrected-pair scope and every prior failed preclaim receipt; no process is killed or launched directly.",{"version":446,"date":383,"sections":447},"20.0.35",[448],{"name":121,"changes":449},[450,451],"Qualify only the existing macOS \u002Ftmp to \u002Fprivate\u002Ftmp system alias for the exact immutable trainer arbitration lock. Preserve lock identity, authority bytes and all other no-symlink\u002Fpath protections. Actual replacement preflight stopped before any helper retirement or successor creation.","Continue the same explicitly approved unused source33 helper replacement scope; no second successor, expanded budget or learner\u002Fdata action follows from this source release.",{"version":453,"date":383,"sections":454},"20.0.34",[455],{"name":121,"changes":456},[457,458],"Import the existing primitive serializer for native selector storage-binding capture. The actual admission review exposed the missing import before any declaration, job change or production image copy.","The human approved replacement of only the two unused release33 migration-control capsules, after native ownership and non-consumer checks, within the unchanged preparation budget. Preserve learner saves, original image and all receipts\u002Ffailures; source correction and native replacement remain separate stages.",{"version":460,"date":383,"sections":461},"20.0.33",[462],{"name":121,"changes":463},[464,465],"Interpret an absent APFSSnapshot field only on a native ExFAT filesystem as not applicable and not a snapshot. APFS missing\u002Funknown flags and explicit ExFAT null\u002Ftrue values remain unknown or held. The actual Extreme preparation preflight exposed this representation mismatch before any preparation claim or image mutation.","Preserve the approved source32 pending archive exception and all original failures; this source correction is separate from installation, image movement and learner activation.",{"version":467,"date":383,"sections":468},"20.0.32",[469],{"name":121,"changes":470},[471,472,473],"Add an explicit human-directed live-rollout source policy that records incomplete full-project qualification honestly and binds focused migration checks, exact source and immutable authorization. The normal full-qualification gate remains unchanged.","Prioritize moving complete saved training state to Extreme through the existing native and managed controls. Data custody, capacity, complete saves, rollback and historical failure evidence remain protected; the human-approved exact pending archive exception preserves its data, intent, failure, hold and full obligation without retry or completion.","Isolated source only; installation, drive movement, publication and optimizer activation remain separate unfinished stages.",{"version":475,"date":383,"sections":476},"20.0.31",[477],{"name":121,"changes":478},[479,480],"Preserve the existing pin\u002Fcollector transaction before native role authority reads; defer only typed native StoreBusy in optional checkpoint collection instead of creating a retention integrity hold.","Keep local admission closure first and all model\u002FAdam\u002Fmemory, reference, custody, capacity and unknown-I\u002FO protections intact. The source update is isolated and has not been installed or activated; both original learners completed their managed final saves and stopped on Samsung.",{"version":482,"date":383,"sections":483},"20.0.30",[484],{"name":121,"changes":485},[486,487],"Keep managed storage role authority reads nonblocking and subject to existing backup deadlines. Preserve exact local closure, paired native completion and custody checks; integrity and unknown I\u002FO errors never become contention retries.","Bind the backup admission deadline implementation into the exact migration source inventory. Production storage, learner release and all prior incident holds remain unchanged while full qualification is pending.",{"version":489,"date":383,"sections":490},"20.0.29",[491],{"name":121,"changes":492},[493,494,495,496],"Add bounded native writer quiescence, persistent outside-image bootstrap\u002Fadmission guards, guarded compatible-control installation and exact existing-job observations. Preserve original complete-save admission while pre-seal work drains; close fresh work before native detach.","Add one-owner artifact stage and final paired completion producers. Effective routing requires the durable completion outcome; migrated historical sealed allocations retain full future obligations unless portability is independently proven. Unknown and partial outcomes remain held without replay.","Bind all new control\u002Fguard\u002Freader sources into the exact migration review. Preserve genuine historical failed-only retention evidence across authenticated14.x schema drift; current-source configuration drift remains held.","Source integration remains isolated. Native orchestration\u002FCLI, final ordinary-role release and full qualification are pending; no production move, limit apply, installation, publication, activation or main push is claimed.",{"version":498,"date":383,"sections":499},"20.0.28",[500],{"name":121,"changes":501},[502,503,504],"Add an exact live pre-seal review and one-use shared managed pause request for each game, preserving committed\u002Funretained work, old-route final-save admission, original quotas and all integrity holds.","Gate integrated coordinator launches, fresh rollouts and new optional writers during the shared sealing phase; let already admitted work drain naturally.","Bind the later paused capture to the same preserved live prefix, owners\u002Fsource, native pause requests and private claim. Partial, foreign, missing or unsafe evidence holds without replay or image rights. Full writer quiescence, native stages, bootstrap and routing remain pending.",{"version":506,"date":383,"sections":507},"20.0.27",[508],{"name":121,"changes":509},[510,511,512],"Bind one exact private cutover journal\u002Fheader\u002Fhead and one-use publication claim to a shared native resource intent under fresh two-game capture and endpoint checks. Keep quotas, immutable authorities and learner ledgers unchanged.","Preserve failed pre-commit claims and reject same\u002Falternate pending requests; inspect shared\u002Flocal mismatches without repair or native stage rights. Keep both effective readers held even for malformed declared intents.","Retain control descriptors and source\u002Ftarget\u002Ftrainer custody checks through the shared publication endpoint. Managed full-writer quiescence, native cutover, final paired completion and all-writer routing remain pending.",{"version":514,"date":383,"sections":515},"20.0.26",[516],{"name":121,"changes":517},[518,519,520],"Add a private local cutover journal with exclusive creation, immutable hash-chained stage intent\u002Foutcome records, fsync and a sealed one-process owner lock.","Inspect partial, torn, reordered, foreign and failed outcomes without repair or replay rights. Keep native route\u002Flearner authority separate from local journal provenance.","Bind the journal implementation into exact migration review source custody. Shared resource fencing, managed seals\u002Ffull writer quiescence, native application and routing remain pending.",{"version":522,"date":383,"sections":523},"20.0.25",[524],{"name":121,"changes":525},[526,527,528],"Add a read-only paired pre-copy migration review binding paused durable prefixes, complete saved component bytes and memory roster, current owners\u002Fsource capsules, pins\u002Freferences\u002Fholds, native resource obligations, selectors\u002Flauncher histories, qualified filesystem mechanisms and exact target custody.","Add non-creating campaign readers for migration consumer inventory; reject missing historical roots, nested target mounts and unsafe active trainer exclusions. Preserve disabled workers' saved databases and markers together.","Reviews grant no image, quota, source-handoff or learner rights. Shared journal\u002Fapplication\u002Fcrash recovery and complete writer\u002Fbootstrap\u002Fcleanup routing remain pending; production stays unchanged.",{"version":530,"date":383,"sections":531},"20.0.24",[532],{"name":121,"changes":533},[534,535,536],"Resolve APFS container growth's native remount only for the exact owned image, unchanged UUID\u002Finode and precise geometry; retain complete payload verification and normal detach.","Add one-use, saved-review post-growth fixture completion preserving all original receipts, copy, resize intent and failures. Remaining rights are copied-image normal detach and final independent offline verification only; no creation, data-test, copy or resize replay.","Production migration, installation, learner activation and main push remain separate pending stages.",{"version":538,"date":383,"sections":539},"20.0.23",[540],{"name":121,"changes":541},[542,543],"Implement bounded native post-data fixture context acquisition and one-use exact saved-review application. Preserve the original account and both failures, full retained payload, current UUID\u002Fancestry and all completed checks; reject open descriptors and changed source\u002Freceipts before tools.","Continue only normal detach, exclusive same-source copy and copied-image resize under existing budgets. Never recreate the fixture, replay completed data tests, force detach or retry a failed application. Production migration and learner activation remain separate pending stages.",{"version":545,"date":383,"sections":546},"20.0.22",[547],{"name":121,"changes":548},[549,550],"Add a pure exact post-data native fixture review gate binding completed checks, both failure\u002Fsource chains, current attachment\u002Fpayload and absence of open writers.","Stage rights exclude source creation and repeating measured data checks. Native context loading\u002Fapplication remains pending; no production storage or training change.",{"version":552,"date":383,"sections":553},"20.0.21",[554],{"name":121,"changes":555},[556,557],"Explicitly close both owned SQLite qualification connections before requesting normal image detach. Transaction contexts alone do not close database handles.","Preserve four passed native data checks and the failed busy-detach outcome. Any continuation still requires exact retained-fixture review; no force detach, replay, production move or learner activation.",{"version":559,"date":383,"sections":560},"20.0.20",[561],{"name":121,"changes":562},[563,564],"Keep image payloads on the approved backing while placing only empty verification mount-control directories outside a mounted disk image, avoiding the observed macOS mount-path permission restriction.","Preserve exact failed native fixture receipts and allow only a source\u002Fcustody-bound reviewed continuation from their pre-data boundary; capture native tool stderr and never recreate, replace or blindly retry a failed fixture.",{"version":566,"date":567,"sections":568},"20.0.19","2026-10-11",[569],{"name":121,"changes":570},[571,572],"Add one-use copied-image resize with readonly pre-resize APFS UUID\u002Fcontent census, exact image\u002Fcontainer growth and independent post-resize UUID, inode, geometry and content verification.","Retain exclusive intent and partial native outcomes. No forced detach, automatic retry, capacity grant or learner rights. Native filesystem qualification and production migration remain separate stages.",{"version":574,"date":567,"sections":575},"20.0.18",[576],{"name":121,"changes":577},[578,579],"Add descriptor-bound detached sparsebundle inventory and exclusive image copying with independent complete destination hashes, fsync, UUID\u002Fcustody checks and cooperative cancellation.","Preserve failed\u002Fpartial artifacts and the unchanged source. Image evidence grants no capacity, deletion or learner authority. Resize and native APFS qualification remain separate incomplete migration stages.",{"version":581,"date":567,"sections":582},"20.0.17",[583],{"name":121,"changes":584},[585,586,587],"Verify paired durable campaign prefixes, exact per-game apply anchors, private completion hash and current destination ancestry before resolving a same-image backing move.","Keep immutable operations, identity and quota receipts intact; distinguish their historical device anchors from fresh destination custody. Pending, missing, foreign or conflicting proofs hold without fallback.","Route native resource, checkpoint and snapshot admission through the verified effective backing and stronger capacity floor. Preserve all peaks, inflight writes and unverified future obligations. Source changes alone grant no migration, capacity or learner activation.",{"version":589,"date":567,"sections":590},"20.0.16",[591],{"name":121,"changes":592},[593,594],"Add legacy-preserving effective storage readers and explicit holds for every incomplete or unproved backing migration.","A new drive, image size or ledger quota declaration alone grants no route or capacity. Complete shared-transaction qualification and production integration remain pending. No installation, storage move, quota apply, publication or learner activation.",{"version":596,"date":567,"sections":597},"20.0.15",[598],{"name":121,"changes":599},[600,601],"Integrate the last unmerged opt-in learning-experiment branch with current physical coupled learner loading, lock-free retention, atomic status and maintenance\u002Fresource guards preserved.","Experiment adoption remains disabled by default and requires its own exact source-bound review. The approved storage migration adopts no experiment and changes no live learning settings.",{"version":603,"date":567,"sections":604},"20.0.14",[605],{"name":121,"changes":606},[607,608],"Reconcile the maintenance-wait-validation branch with bounded exact predicate diagnostics while retaining newer current-owner, schema and ordered-wait safeguards.","Preserve resumable controls and final exact-apply fences. This source-only integration does not install, publish, move storage, amend quota or activate a learner.",{"version":610,"date":567,"sections":611},"20.0.13",[612],{"name":121,"changes":613},[614,615],"Bind native sealed-allocation admission to exact authoritative campaign selector files and complete launcher\u002Fchannel histories. Routine status-cache refreshes do not invalidate full runtime verification; actual selector\u002Fsource\u002Fcustody changes still hold.","Preserve finite deadlines, full immutable verification, all mutable reservations and physical-space protections. Installation alone does not publish or activate a learner release.",{"version":617,"date":567,"sections":618},"20.0.12",[619],{"name":121,"changes":620},[621,622],"Native sealed-allocation reconciliation waits for typed campaign\u002Fresource lock acquisition conflicts within its existing finite deadline, releases partial exclusive claims before waiting, and retains completed full runtime verification. Body failures, custody changes, storage faults and capacity failures remain terminal.","Report native transaction waits separately from runtime I\u002FO; preserve the historical quota, recovery, maintenance and validation evidence. This control installation does not publish or activate a learner release.",{"version":624,"date":567,"sections":625},"20.0.11",[626],{"name":121,"changes":627},[628,629],"Separate bounded native launcher reads from the small reconciliation receipt limit: verify embedded manifests up to 16 MiB per control file and 512 MiB across the declared inventory.","Preserve the complete launcher history, exact source\u002Fcustody checks, the 1 MiB receipt bound, all storage protections, quotas and learners. The first 20.0.10 batch held before mutation; see [the release record](docs\u002Ftraining-20.0.11.md).",{"version":631,"date":567,"sections":632},"20.0.7",[633],{"name":121,"changes":634},[635,636,637],"Reuse one fresh fenced capacity inventory when preferred two-rollout credits fall back to one complete rollout; preserve save margin, inflight growth, host floor and authority\u002FABA checks.","Report native capacity wait reasons and the latest valid check separately from actual trainer failures; preserve exact incident identity and maintenance holds.","Preserve learner, Adam and auxiliary-memory lineage. Source, installation, publication, activation and live throughput are verified separately.",{"version":639,"date":640,"sections":641},"20.0.1","2026-10-09",[642],{"name":121,"changes":643},[644,645,646],"Retry transient campaign\u002Fstorage lock contention without forcing another fresh capacity inventory; retain ordinary expiration, growth, revocation, and atomic write admission.","Retry contention after five seconds while keeping snapshot\u002Fcapacity waits at sixty seconds and limiting wait telemetry to once per minute.","Preserve model\u002FAdam lineage and behavior configuration. Source validation and managed deployment evidence are recorded in [the release record](docs\u002Ftraining-20.0.1.md).",{"version":648,"date":649,"sections":650},"20.0.0","2026-10-08",[651],{"name":121,"changes":652},[653,654,655],"Add an exact reviewed Red41→45GiB recovery control protecting Crystal’s existing28GiB allocation, all retained data and native capacity limits.","Prepare a sealed campaign capsule once and reuse its exact verified source, runtime, configuration and campaign binding for managed publication.","This is isolated recovery preparation, not installation, publication, activation or measured training improvement. See [the release record](docs\u002Ftraining-20.0.0.md).",{"version":657,"date":649,"sections":658},"19.1.53",[659],{"name":121,"changes":660},[661,662],"Keep runtime files smaller than 4 KiB as complete independent copies instead of performing per-file APFS reuse; preserve donor checks and full capsule sealing and verification.","Qualify the same-device clone path with bounded matched fixtures. This is a source-only capsule-build optimization, not live training-throughput evidence. See [the release record](docs\u002Ftraining-19.1.53.md).",{"version":664,"date":665,"sections":666},"19.1.51","2026-10-07",[667],{"name":121,"changes":668},[669,670,671,672],"Integrate reviewed source49\u002F50 quota and same-image storage identity with the final adoption race fix; preserve source-only status and all existing release, capacity and maintenance gates.","Credit the exact preparation-owned Red Mart entrance in the same action, with passive native input timing, and forward configured terminal battle trace limits in managed campaigns.","Make legacy R2 age inventory intrinsically report-only while retaining current guarded cleanup and the weekly catalog.","Preserve and triage outstanding drafts; exclude invalid bounded experiments and superseded runtime reporting. See [the integration record](docs\u002Ftraining-19.1.51.md).",{"version":674,"date":665,"sections":675},"19.1.47",[676],{"name":121,"changes":677},[678,679],"Bind evaluation jobs using enqueue records from the verified ledger, avoiding a full detached history copy per job during status publication. Preserve fresh ledger verification and exact job hashes.","Keep this source-only performance fix separate from production activation and measured live throughput. See [the release record](docs\u002Ftraining-19.1.47.md).",{"version":681,"date":682,"sections":683},"19.1.30","2026-10-06",[684],{"name":121,"changes":685},[686,687],"Reuse already admitted campaign write credits for up to 15 minutes instead of rescanning after 60 seconds. Exhaustion, snapshot growth, forced checks and restart still require fresh admission.","Check physical free space against all checkpoint\u002Fwrite reserves and sealed peaks before each credit debit, without a tree walk. Preserve existing byte limits, quotas, weights and optimizer lineage. See [the release record](docs\u002Ftraining-19.1.30.md).",{"version":689,"date":682,"sections":690},"19.1.29",[691],{"name":121,"changes":692},[693,694,695],"Reuse conservative progressive capacity admission for new reservations, avoiding unconditional scans of every retained runtime while protecting sealed peaks.","Cache bounded, identity-validated capsule inventory path ordering across checks; always measure current file sizes and host free space.","Add a 4096-file sample before full runtime traversal, retaining exact fallback and all quota\u002Ffloor checks. See [the release record](docs\u002Ftraining-19.1.29.md).",{"version":697,"date":682,"sections":698},"19.1.28",[699],{"name":121,"changes":700},[701,702,703],"Add a separately reviewed Red 36-to-40 GiB headroom extension, preserving the earlier quota receipt and compatibility with resident older capsules.","Retain fresh host-floor, checkpoint, writer and sealed-peak admission; bind application to exact source, owner, authority and prior approval evidence.","Record storage timing evidence and distinct source, installation, quota and activation stages in [the release record](docs\u002Ftraining-19.1.28.md).",{"version":705,"date":682,"sections":706},"19.1.27",[707],{"name":121,"changes":708},[709,710,711,712],"Recognize an enemy faint during pending trainer-finisher turn settlement as a deferred native transition instead of falsely aborting the episode with `battle_safety_switch_failed`. No new attack intent or finisher attempt is consumed.","Add state and native-input regressions while preserving genuine timeout and wrong-move safety checks. Exact incident replay was unavailable; native qualification reproduces the same transition class.","Let a storage-deferred or paused evaluation lane yield to other fair queue groups. Preserve capacity admission, the single host evaluator, immutable outcomes and cursor advancement only after dispatch.","Source, installation, managed publication and activation are recorded separately in [the release record](docs\u002Ftraining-19.1.27.md).",{"version":714,"date":665,"sections":715},"19.1.46",[716],{"name":121,"changes":717},[718,719],"Add capture-yielding exact maintenance application: verify outside coordinator locks, commit through nonblocking native exclusion, and release only the reviewed reservation. Never invoke general reconciliation or unlink payloads.","Persist an exact release intent before resource publication and clear the failure hold only with a final review record. Interrupted repairs require fresh custody and matching authority. No production application or deployment is included.",{"version":721,"date":682,"sections":722},"19.1.45",[723],{"name":121,"changes":724},[725,726],"Add explicit bounded, read-only maintenance-failure preparation with detached ledger replay and checkpoint-capture deferral. Accept only ordinary optimizer-progress appends between independently verified snapshots.","Preserve exact native repair proposals and custody checks; no deployment, hold clearing, quota change or deletion is performed by preparation.",{"version":728,"date":682,"sections":729},"19.1.44",[730],{"name":121,"changes":731},[732,733],"Defer routine campaign source upgrades until the active release has six valid completed training attempts per starter. Keep learning while collecting evidence; count across restarts and reject censored, invalid, duplicate or misbound records.","Bound local episode-index reads before release preflight, expose remaining starter counts, and require an exact source\u002Fcandidate-bound reviewed exception for urgent repairs. Preserve integrity holds and managed handoff checks.",{"version":735,"date":682,"sections":736},"19.1.43",[737],{"name":121,"changes":738},[739],"Preserve imported fixed, qualified and explicitly pinned references in cold-storage exclusions even when they are absent from ordinary learner publication history. Report their protection instead of crashing; no retirement authority is added.",{"version":741,"date":682,"sections":742},"19.1.42",[743],{"name":121,"changes":744},[745,746,747],"Bind the cold-reader capability during native capsule construction and managed publication; verify the source contract before allocation. A sealed capability tag grants no qualification or retirement authority.","Add the explicitly approved 12 GiB offline qualification budget, preserving the 2 GiB default. Account every immutable input and object index, enforce category ceilings and preserve a 10 GiB physical floor.","Bind the reviewed budget through native restored execution and evidence validation; reject changed profiles, missing charges and oversized phases. No production quota increase, backup renewal, deployment or cleanup.",{"version":749,"date":750,"sections":751},"19.1.41","2026-10-05",[752],{"name":121,"changes":753},[754,755,756],"Release the native coordinator during checkpoint capacity inventories; serialize host accounting and revalidate meaningful campaign state and current disk headroom before committing. Defer on contention without nested lock waits; retain exact interrupted quota replay.","Report sealed saved-state and learned-memory growth separately from storage growth. Forecast maintenance before next-save and rollout headroom is exhausted; storage growth alone is not a failure or deletion permit.","Import the coordinated finisher test-only tracker comparison repair. No production quota change, cleanup, installation or activation.",{"version":758,"date":750,"sections":759},"19.1.40",[760],{"name":121,"changes":761},[762,763,764],"Admit the full native evaluator peak through resolved bundle paths before starting its payload writes; preserve the existing 2 GiB aggregate ceiling and exact unmet requirements.","Enforce evaluation deadlines with a kernel timer in the owned guarded child. Expiry releases native flocks through process teardown while retaining incomplete one-use evidence and reservations.","Bind accepted evaluation jobs to exact native qualification rules, seeds, cohorts, scenarios, relocated configuration and asset bytes. Hold final aggregate publication failures even when callers catch them.",{"version":766,"date":750,"sections":767},"19.1.39",[768],{"name":121,"changes":769},[770,771],"Rebind restored native evidence to the exact active capsule through managed source ancestry, preserving the original authority and ledger prefix. Verify current frozen assets and unchanged behavior, all twelve sealed phase results, aggregate\u002FRSS accounting and native evaluator exclusions.","Reject fixture domains\u002Fjobs and missing or changed provenance. Evidence validation grants no activation, backup window or production custody authority.",{"version":773,"date":750,"sections":774},"19.1.38",[775],{"name":121,"changes":776},[777,778],"Execute seven native restored reader classes in twelve guarded sequential phases, preserving complete coupled memory, Adam state, stacked observations and fixed-reference evaluation.","Seal results before exact disposable cleanup, record peak RSS, and report exact capacity shortfalls. Hold native evaluator exclusion only during evaluation to avoid backup self-contention.",{"version":780,"date":750,"sections":781},"19.1.37",[782],{"name":121,"changes":783},[784,785,786],"Add a read-only file object provider whose complete immutable input bytes and index memory are included in qualification admission, with exact original deadlines and no network or upload capability.","Measure native campaign charges before and after private candidate quarantine under the operation's exclusive collection lock, preserving zero reclamation and exact quota evidence.","Place guarded native SQLite restore scratch in the existing admitted phase, retaining failed reads and capacity failures for exact custody review and cleanup.",{"version":788,"date":750,"sections":789},"19.1.36",[790],{"name":121,"changes":791},[792,793,794],"Route registered restored evaluation jobs through their admitted private phases, binding original epoch, prospective capsule, exact candidate and fixed reference. Preserve native input, result and one-use attempt checks.","Require reader qualification to include the campaign's exact fixed reference and both native result roles; reject candidate-only evidence for fixed-reference campaigns.","Keep restored fixture execution separate and all live storage\u002Fdeployment gates closed; full restored producer integration remains outstanding.",{"version":796,"date":750,"sections":797},"19.1.35",[798],{"name":121,"changes":799},[800,801,802],"Add disabled exact cold-duplicate quarantine\u002Fremoval transactions with separate current action authority, bounded durable journals and native exclusions. No scheduler or standing policy invokes them.","Preserve failed custody and reservations, support separately reviewed rollback or complete remote restore, and recognize exact interrupted publication without clearing historical holds.","Distinguish gross removed payload bytes from net retained quota charges and physical free space. Busy coordinator admission refuses before authority reads; checkpoint priority is rechecked immediately before mutation.",{"version":804,"date":750,"sections":805},"19.1.34",[806],{"name":121,"changes":807},[808,809,810],"Verify original runtime capsules as complete archive streams through the native capsule inventory seal, retaining only bounded manifests when explicitly selected.","Let the guarded private native backup endpoint read those exact original archives and reuse existing verified objects through native read-before-create deduplication. Original source paths and generation records remain intact; all new writes stay in admitted private phase storage.","Bound private object writes and inventory working memory, and retain failures even when callers catch read errors. Full restored qualification and live retirement remain unimplemented and disabled.",{"version":812,"date":750,"sections":813},"19.1.33",[814],{"name":121,"changes":815},[816,817,818],"Bind restored exact-reference reads to an admitted private native phase, retaining remote copies and reservations until sealed cleanup. Reader failures hold both even when caught by a caller.","Add journaled staging of the guarded private candidate only, preserving all protected-reference exclusions and recording zero reclamation.","Translate explicit asset fields and preserve native reset\u002Fmodel contracts; permit exact journaled cleanup of native read-only disposable directories. Full restored qualification and production retirement remain disabled.",{"version":820,"date":750,"sections":821},"19.1.32",[822],{"name":121,"changes":823},[824,825,826],"Bind restored execution to a one-use guarded child with exact original ledger and protected-state checks. Fixture evidence remains distinct.","Translate only declared private native resource roots while preserving quotas, checkpoint reserves and the physical host floor.","Account native campaign-scoped disposable phases once under the shared aggregate ceiling; no live qualification or retirement authority is enabled.",{"version":828,"date":750,"sections":829},"19.1.31",[830],{"name":121,"changes":831},[832,833,834],"Verify complete native archive streams while materializing only a declared input closure, preserving unselected history coverage and failed phase custody.","Add restored-input provenance preparation and a tested macOS child write guard; keep fixture domains distinct and all native qualification\u002Fretirement gates closed pending producer integration.","Charge retained private campaign\u002Fhost metadata explicitly across consumer phases. No installation, activation or production reclamation is performed.",{"version":681,"date":750,"sections":836},[837],{"name":121,"changes":838},[839,840],"Add an explicitly owned qualification workspace that accounts immutable inputs, disposable outputs, reserved working memory and outstanding reservations under one aggregate ceiling and absolute deadline.","Seal phase results before exact disposable cleanup; preserve failed custody, open-reader outputs and one-use journals. This component grants no production movement, retirement or activation authority.",{"version":689,"date":750,"sections":842},[843],{"name":121,"changes":844},[845,846],"Integrate the preserved retention readers, backup preparation retries and diagnostic maintenance lease in an isolated source candidate, keeping native resolver and deadline protections together.","Preserve the exact finisher26 behavior while combining storage fixes; qualification remains required before any installation or deployment. No live authority or backup window is granted.",{"version":697,"date":750,"sections":848},[849],{"name":121,"changes":850},[851,852],"Compare complete native SQLite memory in bounded chunks under one deadline, with immutable input and complete typed-row evidence.","Keep owned-fixture qualification, quota limits and production activation boundaries unchanged.",{"version":705,"date":750,"sections":854},[855],{"name":121,"changes":856},[857,858],"Match native cold-reader continuation\u002Fresident observation stacking to the resolved training configuration and bind the wrapper source in its evidence.","Preserve exact saved policy\u002FAdam contracts; qualify the repair with a real four-frame recurrent fixture. This source-only increment grants no live retention, deletion, backup, or deployment authority.",{"version":860,"date":750,"sections":861},"19.1.26",[862],{"name":121,"changes":863},[864,865,866,867,868],"Split pending trainer-finisher turn settlement from move submission, preserving the one-use attempt and rechecking eligibility at the command-menu boundary.","Yield before mandatory replacement input when a pending turn faints the active; keep living teammates available through the existing forced-switch controller.","Require fresh used-move evidence when PP has not changed; an enemy-only KO cannot attribute the previous move to a queued finisher. Preserve real PP-based and fresh native move mismatch holds.","Add Red native input-derived pending Growl qualification and wrong-move regressions, plus cross-game execution checks. The exact live incident savestate was unavailable.","Record the separately approved Red-only integration, installation, single candidate publication and exact managed epoch restart in [the rollout record](docs\u002Ftraining-19.1.26-red-rollout.md). Native reconciliation charges the approved 55,296 lost steps; 64 new-source PPO updates and a normal complete checkpoint at 30,501,888 committed steps are verified. A subsequent managed quota wait blocks further learning. Crystal's 19.1.26 handoff was not performed.",{"version":870,"date":750,"sections":871},"19.1.25",[872],{"name":121,"changes":873},[874,875,876],"Add an explicitly reviewed one-use native diagnostic maintenance lease for both resident campaign trainers. Finish updates and required saves before parking; retain learner, Adam, memory, environment, recurrent state and RNG in the same execution. Admission expires after120seconds, and the common parking deadline cannot extend past60seconds after the first acknowledgement.","Defer optional evaluation service, archive dispatch, retention and advisory storage scans while mandatory owner, authority, integrity and checkpoint work keeps priority. Native local custody\u002Fpin\u002Freceipt reads require exact two-game acknowledgements, kernel-held resident presence and drained optional workers; no backup, custody migration or automatic repeat is granted.","Source release only: no installation, activation, scheduler change or live lease. A future managed source handoff preserves saved learner lineage but censors\u002Fresets episodes; total training interruption remains uncertain.",{"version":878,"date":879,"sections":880},"19.1.24","2026-10-04",[881],{"name":121,"changes":882},[883,884],"Add an explicit persistent backup preparation budget with bounded native transaction retries, capture priority and diagnostic receipts. Preserve exact pending custody and immutable learner lineage; source preparation does not install or activate this release.","Apply the same deadline to native capacity\u002Fcustody scans and bundle hashing, bound authority reads after restore, and recheck prepared-cut admission. Persistent CLI runs require a new private receipt file; status pipes are best effort and cannot delay terminal evidence.",{"version":886,"date":879,"sections":887},"19.1.21",[888],{"name":121,"changes":889},[890],"Isolated version2 reader-evidence schema binds all seven consumer classes to exact source\u002Fobject operation receipts and rejects unknown, fabricated, accounting-unsafe or relabeled local reads. Resolver\u002Fnative producer\u002Fplanner integration remains unfinished; no installation, activation or custody authority.",{"version":892,"date":879,"sections":893},"19.1.20",[894],{"name":121,"changes":895},[896],"Isolated engineering draft bounds cold-history report work and cooperative deadlines, clamps process probes to remaining time, and moves hashing and reader probes outside the final campaign transaction. Native endpoint qualification remains incomplete; this source is uninstalled and unactivated.",{"version":898,"date":879,"sections":899},"19.1.19",[900],{"name":121,"changes":901},[902],"Draft read-only cold-history evidence checks bind distinct remote generations to committed native materialized proofs, freshly verify bounded candidate payloads and recheck complete local custody while retaining protected recovery state. Reader qualification and custody operations remain gated; this isolated source is not installed or activated.",{"version":904,"date":905,"sections":906},"19.1.18","2026-10-03",[907],{"name":121,"changes":908},[909,910],"Add opt-in managed host scratch backups with exact cut\u002Frelease replay and archived control provenance. Native frozen 19.1.15 consumers are qualified using private fixtures.","Prepare explicitly reviewed shared host recovery scratch, preserving campaign checkpoint\u002Fwrite reserves and the host free-space floor. Legacy backup behavior remains the default. This source branch is not installed or activated.",{"version":912,"date":879,"sections":913},"19.1.23",[914],{"name":121,"changes":915},[916,917,918,919,920,921,922,923],"Implement the approved Red and Crystal gym cap contract: allow leader entry at the cap and earned leader levels through badge dialogue, and quarantine preexisting over-cap members in the native PC until the exact badge.","Reserve ordinary trainer XP for every possible participant, including sharing, trade and Lucky Egg modifiers; restore temporary legal members before leader entry after native trainer obligations clear.","Add bounded current-box deposit\u002Fretrieval with immutable identity\u002FXP receipts, interruption reconciliation and episode-only preparation stops. Preserve retirement even when the ROM heals a stored member.","Decode Crystal full trainer rosters read-only from pinned native structs. Qualify the retained first-gym PC approaches and native round trips; later routes remain bounded unsupported outcomes.","Verify Crystal's transient native withdrawal copy before inferring retirement, preserve living HP on receipts, settle legal delayed level publication within the fixed XP allowance, and retain bounded ownership through the outbound PC aisle.","Settle Red's native XP-before-level publication only under verified ROM call bytes and the live CPU stack, retaining the original trainer allowance and blocking extra XP, unknown ownership and true cap crossings.","Correct the native Rod\u002FAbe defeat-bit roster mapping and independently verify their 164\u002F111 XP awards against pinned ROM data.","Source release only. Installation, managed candidate publication, recovery and live activation are separate and have not occurred.",{"version":925,"date":905,"sections":926},"19.1.16",[927],{"name":121,"changes":928},[929,930,931],"Back up large sealed evaluation results with separate bounded control-file limits, one-file validation and streamed staging outside the training coordinator lock. Preserve exact measured evidence and retry the original frozen recovery cut after interruptions.","Bound metadata staging by the reserved scratch capacity before adding recovery pins; release unused scratch reservations on failures before preparation.","This release repairs the standalone recovery tool. Live learner and archival controller activation remains separately reported at 19.1.15.",{"version":933,"date":905,"sections":934},"19.1.15",[935],{"name":121,"changes":936},[937,938,939],"Add immutable private R2 recovery generations with protected bundle\u002Fruntime\u002Fasset dependency closure, bounded materialized restore verification, and source-specific pins; report backup freshness and seven-day storage balance.","Inventory runtime dependencies conservatively; historical absolute paths and ambiguous preparations remain retained.","Add separately adopted lossless SQLite archival for inactive failed executions, with reserved materialized restoration and byte verification before redundant-original removal. Preserve failure records, holds, protected bundles and learner lineage.",{"version":941,"date":905,"sections":942},"19.1.14",[943,946],{"name":121,"changes":944},[945],"Combine reviewed archival throughput fixes with the bounded managed-release capacity preflight from prepared 19.1.11. Deploy through the normal compatible both-game handoff; optional proactive control remains unapplied.",{"name":947,"changes":948},"Training",[949,950],"Prepare explicit Red proactive maintenance-control admission below the existing archival watermark while a stable healthy owner continues optimizer updates. Bind exact native worker, historical family custody, consumer authority and fresh capacity without modifying learner or quota state.","Retain existing storage-pressure\u002Fbusy admission and reviewed exact-claim, batch and archival retry corrections. Preserve capture priority during control-capsule verification. Preparation only; no installation, integration, control admission or activation is claimed.",{"version":952,"date":905,"sections":953},"19.1.13",[954],{"name":947,"changes":955},[956,957,958],"Permit an exact, reviewed zero-reclaim Red learner-lock deferral to select a newer compatible maintenance control, binding the current quota wait and complete verified SQLite custody.","Recheck worker, owner, source, policy, receipts, original identities and reservation under exclusion before recording the control update. Preserve existing storage-pressure reviews, capture priority and learner state.","Include the reviewed 19.1.12 exact-claim selection, bounded batch and archival retry deadline corrections. Preparation only: no installation or live activation is claimed.",{"version":960,"date":961,"sections":962},"19.1.12","2026-10-02",[963],{"name":947,"changes":964},[965,966,967],"Validate an exact managed SQLite archival claim without repeating the global family search; retain fresh pressure, quota, host-floor and consumer custody checks.","Stop a bounded archival batch before another selection after its time or byte budget is exhausted.","Start the short archival commit retry window after the initial capture and consumer refresh; retain a fixed deadline, nonblocking locks and fresh custody checks.",{"version":969,"date":179,"sections":970},"20.1.10",[971],{"name":121,"changes":972},[973,974],"Add an exact managed reconciliation for Crystal's recorded 20.1.7 evaluation resource-lock incident. Verify the original source, failed owner, complete learner and immutable measured outcomes; account unsaved work once while preserving the incident and paused state.","Keep the existing Red-specific recovery guards and all automatic release gates. The recovery tool does not publish a candidate, resume learning or launch a trainer.",{"version":976,"date":977,"sections":978},"20.1.9","2026-10-17",[979],{"name":121,"changes":980},[981,982,983],"Defer optional evaluation on typed host-resource lock contention instead of creating an unsafe learner integrity hold. Preserve genuine source, custody and filesystem failures, existing holds, queued jobs and immutable measured outcomes.","Connect Crystal's eastern Route 31 PC journey to the existing surveyed Violet approach and advance only a ROM-verified, CPU-owned A-or-B dialogue wait. Preserve level caps, ordinary-trainer XP reserves, native PC receipts and finite route budgets.","Preserve compatible learned weights, Adam state, auxiliary memory, numeric campaign counters and managed source-release activation boundaries. Existing integrity incidents require their own managed review.",{"version":985,"date":977,"sections":986},"20.1.8",[987],{"name":121,"changes":988},[989,990,991,992],"Add an on-demand six-case Forest replay pack with sealed fixtures, five-minute invocation limits, isolated outputs, child watchdogs and managed training-priority admission.","Add opt-in frozen-policy segment diagnostics with explicit fresh recurrent\u002Fcontroller state, exact bundle custody and separate inference accounting; no training, qualification or promotion authority.","Add default-off bounded early Forest stall anchors and disjoint trace windows. Companion persistence requires dedicated optional-storage admission and never spends primary training logging allowance.","Preserve canonical bedroom-to-Elite-Four training, learner\u002FAdam\u002Fmemory lineage, ordinary evaluation queues and live campaign controls. Source release alone does not install, publish or activate these changes.",{"version":994,"date":977,"sections":995},"20.1.7",[996],{"name":121,"changes":997},[998,999,1000],"Count routine release coverage only from the exact active epoch after validating launch provenance, allowing legitimate same-release worker expansions without accepting inconsistent source claims.","Preserve lawful emergency field healing during active Crystal recovery and finite quest attempts across controller interruptions; repair audited shared-floor return boundaries.","Prepare separate development and held-out Red repair diagnostics while preserving immutable attempts, native legality, learned weights, Adam state, memory and managed activation gates.",{"version":1002,"date":195,"sections":1003},"20.1.6",[1004],{"name":121,"changes":1005},[1006,1007],"Add an exact managed review for early admission of an unadmitted, same-release Crystal worker expansion. Bind its native request, active source, preserved learner\u002Freference, runtime, configuration, control and integrity evidence; waive only the per-starter source-run waiting period.","Keep native physical compatibility qualification, fresh throughput baseline, bounded capacity trial, operator precedence and protected-checkpoint qualification unchanged. Preserve learned weights, Adam state, auxiliary memory, historical failures and existing release handoff rules.",{"version":1009,"date":195,"sections":1010},"20.1.5",[1011],{"name":121,"changes":1012},[1013,1014,1015],"Rebase every Crystal projection qualification writer into its owned output namespace before native environment construction.","Add exact reviewed supersession of unadmitted namespace-failed worker expansion, preserving failed evidence and the learner hold while permitting the ordinary compatible source handoff.","Make publication capacity proof fail closed on store contention instead of blocking under the resource lock.",{"version":1017,"date":195,"sections":1018},"20.1.4",[1019],{"name":121,"changes":1020},[1021,1022,1023,1024],"Repair late Crystal Egg-return healing across the native New Bark\u002FRoute29 boundary.","Extend audited postgift navigation through Route 30, Route 31 and the Violet gate while preserving native arrivals, battle ownership and finite rescue budgets.","Keep newer operator stop\u002Fpause requests authoritative during automatic campaign trial review.","Preserve accumulated Red recovery fixes and campaign migration safeguards; source publication and live activation remain separate managed stages.",{"version":1026,"date":209,"sections":1027},"20.1.1",[1028],{"name":121,"changes":1029},[1030],"Add an exact managed Crystal three to six worker migration with immutable learner and fixed-reference preservation and a bounded capacity trial.",{"version":1032,"date":567,"sections":1033},"20.0.10",[1034],{"name":121,"changes":1035},[1036,1037,1038],"Reconcile fully verified private read-only runtime allocations through bounded exact native receipts; unclassified runtimes stay fully charged and never earn sampled or cached admission credit.","Reserve the publisher's complete temporary peak before any capsule copy, enforce finite chunk\u002Fmetadata writes, preserve partial preparations and bind whole completion before sealed accounting.","Preserve nested runtime and script consumers, annotate exact filesystem failures, and keep all mutable quotas, pending peaks, save\u002Fwrite obligations, host floors and independent maintenance holds intact.",{"version":1040,"date":567,"sections":1041},"20.0.9",[1042],{"name":121,"changes":1043},[1044,1045],"Coordinate expensive campaign capacity refreshes with bounded native process ownership, FIFO rollout fairness, ordinary snapshot priority, and urgent stop sealing.","Move native snapshot peak inventory outside exclusive campaign\u002Fresource transactions while preserving fresh capacity, root, resource generation, and complete-update gates.",{"version":1047,"date":567,"sections":1048},"20.0.8",[1049],{"name":121,"changes":1050},[1051],"Preserve safe capacity inventories across verified same-owner operational notices with bounded charged growth, prefix and control fences, and cooperative progress diagnostics.",{"version":1053,"date":1054,"sections":1055},"20.0.6","2026-10-10",[1056],{"name":121,"changes":1057},[1058],"Bind the approved Red 50 GiB quota review to Crystal’s current managed 32 GiB reservation; retain the sealed 45 GiB review’s original 28 GiB binding.",{"version":1060,"date":1054,"sections":1061},"20.0.5",[1062],{"name":121,"changes":1063},[1064],"Add the approved exact Red 45-to-50 GiB quota amendment, preserving receipt ancestry, Crystal allocation, fresh host capacity checks, and learner state. This control release does not publish a learner release.",{"version":1066,"date":1054,"sections":1067},"20.0.4",[1068],{"name":121,"changes":1069},[1070],"Add opt-in process-local resumable Red maintenance verification while preserving exact custody and application fences.",{"version":1072,"date":1054,"sections":1073},"20.0.3",[1074],{"name":121,"changes":1075},[1076],"Permit exact current-owner resource-wait observations during bounded maintenance review while retaining authority, archive, quota and commit fences.",{"version":1078,"date":1054,"sections":1079},"20.0.2",[1080],{"name":121,"changes":1081},[1082],"Execution phase diagnostics and contention repairs. Live activation requires a separate managed handoff.",{"version":1084,"date":665,"sections":1085},"19.1.50",[1086],{"name":121,"changes":1087},[1088],"Add an exact append-only same-image storage-identity adoption that preserves immutable policy\u002Fquota receipts across device renumbering while validating current UUIDs, paths and directory inodes. Require durable managed pause and matching upgraded readers; no live adoption or expansion is implied.",{"version":1090,"date":665,"sections":1091},"19.1.49",[1092],{"name":121,"changes":1093},[1094],"Add an exact reviewed Red40→41GiB quota-control amendment with preserved legacy receipt chains, protected Crystal32GiB headroom and explicitly bound rollout staging capacity. Source\u002Fcontrol build only; no learner publication is implied.",{"version":1096,"date":665,"sections":1097},"19.1.48",[1098,1101],{"name":121,"changes":1099},[1100],"Integrate main 19.1.30, reviewed storage 19.1.46 and measured status-copy optimization 19.1.47 while preserving cooperative reads, cold-context routing and the six-runs-per-starter release gate. Source preparation only; see `docs\u002Ftraining-19.1.48.md`.",{"name":1102,"changes":1103},"Dashboard",[1104,1105,1106,1107],"Sweep R2 directly during major-version archival, with resumable scans, deletion accounting, and protection for unknown legacy provenance and saved evidence.","Collapse archived release dropdown entries into major-version model reviews with lifetime counts, while labeling retained historical detail.","Group release history by major and minor version with lifetime run counts and compact historical model review pages.","Retire eligible telemetry more than two major releases behind through bounded Cloudflare maintenance, preserving counts and protected evidence.",{"version":1109,"date":961,"sections":1110},"19.1.10",[1111],{"name":121,"changes":1112},[1113,1114],"Expose Crystal rescue action\u002Fintervention budgets, deadlines, cooldowns and stop\u002Fyield reasons in step traces and episode diagnostics, including bounded compact failure context.","Retain native Elm scene, item-5 Ball count changes and the episode-local gift receipt beside rescue decisions. Preserve controller behavior, budgets and policy credit. Source only; publication held for coordination.",{"version":1116,"date":961,"sections":1117},"19.1.9",[1118],{"name":121,"changes":1119},[1120],"Allow exact reviewed recovery of source-bound SQLite archives with ordinary maintenance reservations as well as explicitly borrowed reservations. Preserve original and archive bytes, immutable failure history, ownership and source checks, and exact one-use reservation release.",{"version":1122,"date":961,"sections":1123},"19.1.8",[1124],{"name":121,"changes":1125},[1126,1127],"Continue active bounded Red Route1 rescue through the measured Viridian exit, with guarded post-parcel delivery recovery on qualified native edges.","Share two interventions per episode and96 total commands per intervention, including owned waits and bounded blocked-edge retries. Preserve controller ownership, suppressed assisted navigation rewards, model\u002FAdam\u002Fmemory lineage, and separate assisted-progress reporting.",{"version":1129,"date":961,"sections":1130},"19.1.7",[1131],{"name":31,"changes":1132},[1133,1134],"Account for campaign and archive storage before inspecting historical runtimes during tight host capacity checks.","Progress from 64 to 512 current runtime file sizes only as needed to establish conservative headroom without traversing complete historical packages. Preserve exact fallback scans, quota limits, sealed peaks and the 10 GiB host floor.",{"version":1136,"date":961,"sections":1137},"19.1.6",[1138],{"name":31,"changes":1139},[1140,1141],"Add an exact reviewed Red-only quota extension from32to36GiB with fresh native admission, immutable legacy policy recognition, crash receipts and preserved learner state.","Report the reviewed quota ceiling without changing Crystal limits, training rules, rewards or Route1 behavior.",{"version":1143,"date":1144,"sections":1145},"19.1.5","2026-10-01",[1146],{"name":121,"changes":1147},[1148],"Campaign checkpoints use SQLite backup for all SQLite memory participants, avoiding slow APFS clone WAL finalization and cold integrity reads while preserving full checks, optimizer state and the 120 second deadline.",{"version":1150,"date":1144,"sections":1151},"19.1.4",[1152],{"name":121,"changes":1153},[1154,1155],"Record per-database clone, open, private WAL checkpoint, metadata, integrity check, close, fsync and checksum durations in existing snapshot completion\u002Ffailure events. Preserve the 120-second deadline, publication integrity, training behavior and learner state. Installation and managed activation remain separate steps.","Admit exact manual timeout recovery when the native scheduler subsequently reports the exited owner's missing lock. Bind secondary notices into the review, preserve the original incident, reject superseding control or release work, and retain once-only loss accounting.",{"version":1157,"date":1144,"sections":1158},"19.1.3",[1159],{"name":121,"changes":1160},[1161],"Propagate the existing host capture-priority callback through maintenance archive checksum reads, checking before each bounded chunk. Preserve full stored and expanded hashes, archival exclusions and accounting, the 120-second snapshot deadline, and learner\u002FAdam\u002Fmemory state. Installation and managed activation are separate steps.",{"version":1163,"date":1144,"sections":1164},"19.1.2",[1165],{"name":121,"changes":1166},[1167,1168],"Make shared gym preparation yield on incomplete or invalid native party XP evidence before species lookup. Prevent the species-zero worker crash and refuse controller inputs or headroom certification until evidence is complete. Preserve policy and optimizer contracts; installation and activation are separate managed steps.","Add exact manual review-bound reconciliation for the demonstrated native species-zero worker exception. Preserve its incident and paused hold, bind console\u002Fsource\u002Flease\u002Fsnapshot evidence, and account approved loss once before the managed fixed-source handoff.",{"version":1170,"date":1144,"sections":1171},"19.1.1",[1172],{"name":121,"changes":1173},[1174],"Make Crystal’s audited Mystery Egg return proactive and bounded, with native Elm interaction and verified five-Poké-Ball gift required before route advancement. Preserve policy action\u002Fobservation contracts and campaign learning state.",{"version":1176,"date":1177,"sections":1178},"19.1.0","2026-09-30",[1179],{"name":121,"changes":1180},[1181],"Add shared cap-relative gym preparation for Red and Crystal, with safe native XP rotation and bounded retries. Installation and live activation remain pending validation.",{"version":1183,"date":1177,"sections":1184},"19.0.39",[1185],{"name":121,"changes":1186},[1187],"Allow an exact existing archival peak to adopt guarded unused-write-credit borrowing when quota tightens after first admission. Keep snapshot, margin and in-flight credits protected; block new writes until lossless archival releases the peak.",{"version":1189,"date":1177,"sections":1190},"19.0.38",[1191],{"name":121,"changes":1192},[1193],"Retry transient historical SQLite commit-lock contention for up to five seconds, releasing all learner locks between attempts and revalidating consumer bindings. Preserve archive custody on timeout; no training, quota or capture-deadline change.",{"version":1195,"date":1177,"sections":1196},"19.0.37",[1197],{"name":1198,"changes":1199},"Fixed",[1200],"Review a later stale maintenance failure for an already abandoned SQLite task by binding the original durable review and revalidating full original\u002Farchive custody. Require its released peak to remain absent; append only the new failure acknowledgment, preserving the original abandonment, all files, learner state, quota and retention. Recheck exact ledger\u002Ffile\u002Fresource bindings under managed exclusion before apply.",{"version":1202,"date":1177,"sections":1203},"19.0.36",[1204],{"name":1198,"changes":1205},[1206,1207,1208,1209],"Use fresh, nofollow large-file metadata as a conservative lower bound for retained runtime reservations. Manifest sizes choose files only; actual inode sizes grant capacity. Retain full selected-campaign quota measurement and exact fallback when the host bound is inconclusive, with the existing resource lock, peaks, 32 GiB ceiling and 10 GiB floor.","Check the combined dual-game release staging overlap with authoritative capacity bounds instead of walking all retained host roots twice for advisory inventories.","Capture up to eight independent SQLite participants concurrently under the same complete snapshot barrier and deadline, avoiding three queued waves for Red's seven participants. Preserve integrity checks, deterministic manifests and failed-save holds.","Publish owner-bound status after completed optimizer updates, at most once per minute without storage scans; refresh coordinator commit observations after slow inventory so they cannot overwrite progress with an earlier heartbeat.",{"version":1211,"date":1212,"sections":1213},"19.0.35","2026-09-29",[1214],{"name":1198,"changes":1215},[1216,1217],"Add an exact managed review for a verified but unfinished SQLite archive whose source epoch changed. Preserve original databases, published archive and receipt, failed worker evidence, and learner lineage while recording abandonment before releasing only the matching borrowed maintenance peak.","Skip that historical SQLite base in later archive selection; fail closed on changed bytes, authority, worker state or peak identity. Keep quota and host-floor admission unchanged.",{"version":1219,"date":1212,"sections":1220},"19.0.34",[1221],{"name":1198,"changes":1222},[1223,1224],"Bound exact host storage inventory to four concurrent independent reservation-root scans when the conservative capacity bound cannot decide. Keep the selected learner's quota scan, exact inode accounting, existing resource lock, 32 GiB per-game ceiling, and 10 GiB host floor.","Start the one-use maintenance worker claim's 600-second lease after slow eligibility and capsule checks, so preflight cannot consume its validity window. Preserve existing failure and review holds.",{"version":1226,"date":1212,"sections":1227},"19.0.33",[1228],{"name":1198,"changes":1229},[1230,1231],"Wait on a borrowed maintenance write allowance before repeating capacity scans; recheck full admission after native maintenance releases the reservation.","Preserve atomic write-credit checks, existing rollout grants, resource limits, learner weights and Adam state.",{"version":1233,"date":1234,"sections":1235},"19.0.31","2026-09-28",[1236],{"name":121,"changes":1237},[1238],"Bound superseded source-boundary recovery pins through explicit native retention, reuse identical runtime pages in new APFS capsules, and replace repetitive AI storage audits with a bounded native monitor. Preserve best\u002Freference weights, current recovery, failed-source evidence, optimizer lineage and immutable capsules. See `docs\u002Ftraining-19.0.31.md`.",{"version":1240,"date":1234,"sections":1241},"19.0.30",[1242],{"name":121,"changes":1243},[1244],"Gate maintenance borrowing on the resident learner's write-credit protocol, reconcile completed SQLite scratch reservations directly in managed dispatch, and defer inconclusive open-file probes without creating a maintenance failure hold. Preserve exact archives, learner and optimizer lineage, protected checkpoints, quota, and host floor. See `docs\u002Ftraining-19.0.30.md`.",{"version":1246,"date":1234,"sections":1247},"19.0.29",[1248],{"name":121,"changes":1249},[1250],"Treat a timed-out SQLite open-file probe as inconclusive and keep its source untouched. Allow exact review of the recorded Red timeout only after the same immutable family has a later verified archive receipt. Preserve the failed event, quota, learner weights, optimizer state and protected checkpoint. See `docs\u002Ftraining-19.0.29.md`.",{"version":1252,"date":1234,"sections":1253},"19.0.28",[1254],{"name":121,"changes":1255},[1256],"Let bounded maintenance staging borrow only unspent write credit when Red reaches its quota, and block new write grants until that exact peak is released. Add an exact reviewed control update after the recorded storage-pressure deferral. Keep the snapshot reserve, in-flight writes, 32 GiB quota, 10 GiB host floor and learner lineage protected. See `docs\u002Ftraining-19.0.28.md`.",{"version":1258,"date":1234,"sections":1259},"19.0.27",[1260],{"name":121,"changes":1261},[1262],"Prevent a completed or unneeded SQLite archive from turning a stale maintenance claim into a lasting hold. Require an exact review of the two existing failures before a newer managed control capsule resumes cleanup. Keep the campaign quota, original evidence, learner weights, optimizer state and source handoff rules unchanged. See `docs\u002Ftraining-19.0.27.md`.",{"version":1264,"date":1234,"sections":1265},"19.0.26",[1266],{"name":121,"changes":1267},[1268],"Add separately approved, lossless archival and bounded restoration of inactive successful source-handoff SQLite families through the existing managed maintenance worker. Preserve all protected bundles, optimizer lineage, quota and host floor. See `docs\u002Ftraining-19.0.26.md`.",{"version":1270,"date":1234,"sections":1271},"19.0.25",[1272],{"name":121,"changes":1273},[1274],"Defer optional evaluators when retaining the next checkpoint would prevent renewal of the snapshot reserve and whole-rollout allowance. Keep current evaluations immutable, actual saves available, and quota\u002Ffloor limits unchanged. See `docs\u002Ftraining-19.0.25.md`.",{"version":1276,"date":1234,"sections":1277},"19.0.24",[1278],{"name":121,"changes":1279},[1280],"Admit one complete bounded rollout when the optional two-rollout credit grant exceeds quota, preserving checkpoint reserve, growth allowance, host floor and managed ownership. See `docs\u002Ftraining-19.0.24.md`.",{"version":1282,"date":1283,"sections":1284},"19.0.23","2026-09-27",[1285],{"name":121,"changes":1286},[1287,1288],"Preserve next-rollout storage headroom against evaluator admission across snapshot-only checks; defer optional work during active snapshots while keeping archive reclamation possible under existing quota and host-floor limits.","Show recorded episode reward and unique positions in the web run diagnostics, and add a Copy JSON action for the exact run record.",{"version":1290,"date":1283,"sections":1291},"19.0.22",[1292],{"name":121,"changes":1293},[1294],"Defer background evaluator discovery when the host evaluator or another campaign transaction is occupied, preserving the fair queue and all launch\u002Fintegrity checks so a slow source handoff does not park the other learner controller. See `docs\u002Ftraining-19.0.22.md`.",{"version":1296,"date":1283,"sections":1297},"19.0.21",[1298],{"name":121,"changes":1299},[1300],"Reject Red learned Forest cross-map recovery edges unless the directional input began at an audited north or south gate approach and reached the matching gate map. Ignore older invalid edges during planning without deleting their evidence; the surveyed recovery route takes over. See `docs\u002Ftraining-19.0.21.md`.",{"version":1302,"date":1283,"sections":1303},"19.0.20",[1304],{"name":121,"changes":1305},[1306],"Fix the campaign scheduler lock inversion that blocked both learners and status publication. Resolve campaign authority before taking the host resource lock, with a bounded scheduler\u002Flearner concurrency regression. Preserve learning rules, optimizer lineage and managed source handoff. See `docs\u002Ftraining-19.0.20.md`.",{"version":1308,"date":1309,"sections":1310},"19.0.19","2026-09-26",[1311,1315],{"name":1198,"changes":1312},[1313,1314],"Close the Red Forest readiness\u002Fretreat gap at row 36, resolve seven Gen I move-name aliases, and count six native direct-damage effects in roster readiness.","Reject Luna capture advice before Poké Ball acquisition and advice that restores Nuzlocke eligibility to retired Pokémon.",{"name":1316,"changes":1317},"Diagnostic",[1318],"Add a default-off, native-proof Crystal rival FIGHT navigation assist for bounded frozen comparisons. Live policy move choice and battle-item legality remain in place.",{"version":1320,"date":1309,"sections":1321},"19.0.18",[1322],{"name":1198,"changes":1323},[1324,1325],"Campaign status scans only the selected campaign's storage instead of every retained runtime capsule. It preserves measured quota usage, peak reservations and physical free space; host-wide availability and forecasts are explicitly unavailable without a full host inventory. Full capacity admission and explicit host inventory retain their existing safety checks.","Scoped storage observations can be reused only while fresh, keep quota-growth forecasts separate, and cannot synthesize host capacity or compare forecasts across inventory scopes. Maintenance peak identity now participates in forecast validity.",{"version":1327,"date":1309,"sections":1328},"19.0.17",[1329],{"name":121,"changes":1330},[1331],"Finalize campaign episode logging on normal and failed exits; close raw log writers even if terminal reporting fails, while preserving the original trainer exception.",{"version":1333,"date":1309,"sections":1334},"19.0.16",[1335],{"name":121,"changes":1336},[1337],"Extend the exact reviewed paused-reconciliation path to clean, unpublished snapshot preflight failures caused by a campaign quota limit. Preserve the last complete bundle and charge any unretained optimizer steps only after an exact operator review. This control-only release does not publish or activate new learner code.",{"version":1339,"date":1309,"sections":1340},"19.0.15",[1341],{"name":121,"changes":1342},[1343,1344,1345,1346],"Keep managed scheduler handoffs pending while an existing job is active, avoiding a competing manual trigger; continue to require an exact completed sealed run before writing its receipt.","Add bounded, sealed candidate decisions to new campaign milestone reports, including cohort coverage and the separate development, qualification, learner, and protected-checkpoint outcomes. Incomplete qualification evidence stays distinct from measured model failure; historical reports remain immutable.","Add an atomic `--review-out` file outside campaign storage for managed reconciliation dry runs, with an apply hint that still requires the exact reviewed record. Show the latest started release handoff and hold-specific next step in campaign status and the dashboard; carry the safe gate and checkpoint fields through cloud publishing.","Expose measured optimizer speed, evaluation queue age, and separate snapshot preflight and barrier times. Reuse the selected campaign's capacity scan during snapshot reservation; keep host-wide checks, checkpoint cadence, and one host evaluator unchanged.",{"version":1348,"date":1309,"sections":1349},"19.0.14",[1350],{"name":121,"changes":1351},[1352],"Recover exact managed scheduler handoff receipts from a verified completed run of the already bound sealed launcher. This control-only release avoids extra scheduler triggers after a slow or racing handoff; learner source epochs, policy weights, Adam state, memory, and protected checkpoints remain unchanged.",{"version":1354,"date":1355,"sections":1356},"19.0.13","2026-09-25",[1357],{"name":121,"changes":1358},[1359],"Add a native-replayed, sparse Red Route 1 north-exit direction field for the exact pre-parcel waypoint 18. Unknown and unreachable tiles fail closed; route observations and high-water progress credit leave the semantic exit proof and learned checkpoint lineage intact. Require an exact resolved-config transform through the managed release publisher.",{"version":1361,"date":1355,"sections":1362},"19.0.12",[1363],{"name":121,"changes":1364},[1365],"Reserve snapshot storage before starting the bounded worker barrier, so the host-wide capacity scan does not consume the admitted 120-second capture and publication deadline. Record preflight and barrier timings separately while retaining exact capacity checks, component verification, and failure holds.",{"version":1367,"date":1355,"sections":1368},"19.0.11",[1369],{"name":121,"changes":1370},[1371,1372,1373,1374],"Finish each independent SQLite snapshot component's backup, integrity check, fsync and hash in the bounded worker pool, preserving manifest order and the publication barrier. Record capture phase timings for successful and failed attempts.","Show exact current and historical failed evaluation gates, unresolved checkpoint failures and unretained optimizer steps in campaign status and the dashboard. Restore missing or replaced report JSON from its immutable ledger event.","Emit durable, deduplicated gate notices to the managed scheduler alert path without stopping the experimental learner or changing the protected checkpoint.","Permit one tightly scoped managed recovery after a newly marked clean unpublished snapshot timeout, with a 24-hour circuit breaker and durable review hold for ambiguous or repeated failures.",{"version":1376,"date":1355,"sections":1377},"19.0.10",[1378],{"name":121,"changes":1379},[1380,1381],"Capture independent quiescent SQLite memories concurrently within the existing worker barrier and 120-second publication deadline. Preserve each full integrity check and component verification.","Classify a SQLite integrity check interrupted by the elapsed capture deadline as a clean deadline failure eligible for the existing single recorded retry; other SQLite errors remain holds.",{"version":1383,"date":1355,"sections":1384},"19.0.9",[1385],{"name":121,"changes":1386},[1387],"Restore one clean, recorded snapshot deadline retry after each completed checkpoint. A second deadline at the same checkpoint or uncertain publication still holds training; 250k checkpoints and the 120-second deadline remain unchanged.",{"version":1389,"date":1355,"sections":1390},"19.0.8",[1391],{"name":121,"changes":1392},[1393,1394,1395],"Normalize frozen evaluation configuration before Crystal asset validation so future sealed development panels can run; preserve failed historical panels without rerolls.","Permit one recorded retry after a snapshot deadline only when the staged bundle is proven unpublished and removed. Reconcile the older exact wrapped publication deadline through the reviewed managed path; a second deadline or uncertain publication retains the hold.","Allow an exact source-epoch report cadence of 7.5M optimizer steps for both campaigns while keeping 250k checkpoints, 1M development checks, and 5M qualification nomination windows.",{"version":1397,"date":1398,"sections":1399},"19.0.7","2026-09-24",[1400],{"name":121,"changes":1401},[1402],"Permit an exact one-use reviewed zero-work or single-uncommitted-rollout restart after a prior paused reconciliation, while binding the already charged historical loss to its immutable receipt and current complete bundle. Missing or changed reconciliation evidence still holds the learner.",{"version":1404,"date":1398,"sections":1405},"19.0.6",[1406],{"name":121,"changes":1407},[1408],"Treat an exact completed paused reconciliation of a failed trainer as the already charged recovery it is. The managed coordinator still requires an explicit resume, valid retained snapshot, matching receipt, and no later owner activity before it can launch ordinary continuation. Unreconciled failures retain the 24-hour recovery limit.",{"version":1410,"date":1398,"sections":1411},"19.0.5",[1412],{"name":121,"changes":1413},[1414],"Add an exact reviewed reconciliation for a trainer that exceeded its pause snapshot deadline after an automatic release handoff. Bind the original failed execution, subsequent release hold and explicit abandonment before charging unsaved work as lost and retaining the last complete learner.",{"version":1416,"date":1398,"sections":1417},"19.0.4",[1418],{"name":121,"changes":1419},[1420,1421],"Skip macOS Metal availability and RNG probes for explicitly CPU configured training and evaluation; retain CPU RNG while ignoring historical MPS RNG during a CPU continuation.","Permit a new exact managed recovery from an admitted, never launched learner after its frozen evaluator is terminal and handled. A pending evaluation or any execution claim still prevents recovery.",{"version":1423,"date":1398,"sections":1424},"19.0.3",[1425],{"name":121,"changes":1426},[1427,1428],"Prime CPU autograd before Crystal environment workers and rollout reservations, so a Metal device-discovery stall cannot charge another update before the optimizer starts.","Add an exact, one-use reviewed recovery for a single collected but uncommitted rollout; preserve its charged and lost counts, complete model\u002FAdam\u002Fmemory bundle, and immutable incident evidence.",{"version":1430,"date":1398,"sections":1431},"19.0.2",[1432],{"name":121,"changes":1433},[1434],"Dispatch a frozen development panel after a bounded trial reaches its clean optimizer cap, while retaining the hold for all other paused learners and later integrity incidents.",{"version":1436,"date":1398,"sections":1437},"19.0.1",[1438],{"name":121,"changes":1439},[1440],"Let an exact reviewed zero-work startup recovery proceed when an older Crystal scheduler controller repeats the same recovery-limit hold after the review; still reject changed incidents, owners, source epochs and snapshots.",{"version":1442,"date":1398,"sections":1443},"19.0.0",[1444],{"name":121,"changes":1445},[1446,1447],"Begin the shared 19.x training source for Red and Crystal with the per-game diagnostic skill scorecards and unchanged checkpoint\u002Fmodel lineage.","Preserve managed campaign source handoff, immutable learner and optimizer bundles, qualification holds, and separate evidence for installation versus live activation.",{"version":1449,"date":1398,"sections":1450},"18.0.40",[1451],{"name":121,"changes":1452},[1453,1454],"Extend the diagnostic skill scorecard to Crystal's three starters and sealed panels, with explicit game identity and missing frozen-cell coverage.","Score native healing attempts separately from recovery resolution, expose Poké Ball access by episode, and count a native zero-win total as zero rather than missing event telemetry. Preserve unknown Crystal recovery rates when archived logs contain heals but no recovery starts.",{"version":1456,"date":1398,"sections":1457},"18.0.39",[1458],{"name":121,"changes":1459},[1460,1461],"Add a read-only Red skill scorecard for sealed evaluation panels and completed training episodes, with per-starter battle, catching, healing, overworld, safety, autonomy, and resource evidence, explicit opportunity counts, and missing-cell coverage.","Record native catch, recovery, and trainer-battle-start opportunities in future semantic evaluation results. Preserve missing evidence on historical panels as unknown, and keep scores separate from rewards, qualification, and protected checkpoint promotion.",{"version":1463,"date":1398,"sections":1464},"18.0.38",[1465],{"name":121,"changes":1466},[1467,1468],"Bind each newly admitted campaign to its existing Red or Crystal OpenClaw job before reporting CLI success. Verify the sealed launcher, preserve schedule and alerts, run a managed tick, and record a durable handoff receipt.","Expose an incomplete scheduler handoff in campaign status and provide an idempotent `bind-scheduler` retry after a partial admission. Keep recovery qualification and optimizer activation separate.",{"version":1470,"date":1471,"sections":1472},"18.0.37","2026-09-23",[1473],{"name":121,"changes":1474},[1475,1476],"Add an opt-in exact-reference actor penalty for a reviewed Red recovery comparison. Preserve critic learning and the protected pointer; bind the reference to verified immutable policy bytes after each managed load.","Admit bounded Red retention trials from an exact saved policy, Adam state, schedule, RNG and auxiliary memory, with fixed training seeds, 250k development checks and a 500,736-transition cap. Keep historical failures and qualification seeds separate.",{"version":1478,"date":1471,"sections":1479},"18.0.36",[1480],{"name":121,"changes":1481},[1482],"Add a one-use reviewed managed recovery for an exact zero-work campaign startup loss after the automatic two-per-24-hour recovery limit. Bind the review to the stopped process, durable hold, immutable learner bundle and source epoch; retain the automatic retry cap and reject stale or ambiguous evidence.",{"version":1484,"date":1471,"sections":1485},"18.0.35",[1486],{"name":121,"changes":1487},[1488,1489],"Restore the audited Red opening route target after an explicitly forgiven rival loss, and add opt-in bounded Route 1 stall assistance from the existing native-surveyed WP16\u002FWP17 fields. Mark assisted actions controller-owned, suppress their navigation credit, and exclude assisted episodes from protected-checkpoint promotion.","Require exact Route 1 objective and field geometry before a managed publisher enables the Red rescue. Keep Crystal behavior and both checkpoint\u002Foptimizer lineages intact.",{"version":1491,"date":1471,"sections":1492},"18.0.34",[1493],{"name":121,"changes":1494},[1495],"Add an exact reviewed reconciliation command for a finished SQLite snapshot-capture trainer exception. Preserve the incident and protected checkpoint, retain the last complete learner\u002FAdam\u002Fmemory bundle, and charge unsaved work as lost before a managed resume.",{"version":1497,"date":1471,"sections":1498},"18.0.33",[1499],{"name":121,"changes":1500},[1501,1502,1503],"Add an explicit, exact Red campaign release option to restore the verified Pewter Gym east-aisle recovery branch to an older resolved configuration. Reject unexpected Gym route shapes and preserve every other behavior setting.","Review the current Red and Crystal campaign failures with bounded episode evidence. Keep Crystal battle rules unchanged pending a controlled Cyndaquil rival intervention test.","Load an attested campaign bootstrap helper without writing bytecode into its sealed runtime capsule; make the dead evaluator recovery test independent of host process-list permissions.",{"version":1505,"date":1471,"sections":1506},"18.0.32",[1507],{"name":121,"changes":1508},[1509,1510,1511],"Add reviewed reconciliation for stopped capacity exceptions and managed cross-volume relocation of paused Red and Crystal campaigns into a dedicated external APFS image.","Preserve exact durable learner policy, Adam state and memory during relocation, archive old immutable campaign history, and retain the protected checkpoint through an exact imported pointer. New external campaigns require a separate managed resume.","Read sealed, no-WAL SQLite reference memory without attempting sidecar writes during cross-volume import, and archive an interrupted unadmitted import through an exact review before retry.",{"version":1513,"date":1471,"sections":1514},"18.0.31",[1515],{"name":121,"changes":1516},[1517,1518,1519],"Add an immutable per-campaign release channel and a one-command publisher for verified Red and Crystal source capsules.","Let the managed coordinator request a cooperative old-epoch pause, verify the exact sealed learner boundary, and activate a compatible release without resetting weights, Adam state, memory or protected checkpoints. Incidents and unsafe or incomplete stops retain a review hold.","Allow a verified control capsule to run the existing scheduler job while the learner remains on its recorded source epoch. Source implementation and live job deployment remain separate stages.",{"version":1521,"date":1471,"sections":1522},"18.0.30",[1523],{"name":121,"changes":1524},[1525,1526],"Retain the latest observed trainer-battle entry party, HP, moves and PP in each Red and Crystal episode, with explicit unknown evidence when the pre-battle snapshot is incomplete.","Count policy, controller and rejected battle steps alongside native turn-resolution and player-move PP evidence for that same trainer battle, and include the bounded record in episode exports. These diagnostics do not alter actions, rewards or learned weights.",{"version":1528,"date":1529,"sections":1530},"18.0.29","2026-09-22",[1531],{"name":121,"changes":1532},[1533,1534],"Prevent Crystal's single starter from earning optional wild-battle XP near the active Johto level cap, while retaining a legal Ball path for eligible catches.","Mask policy A confirmations in the native Crystal battle Pack when no legal catch is available or the selected pocket is not Balls. Preserve the ROM item guard and mark cap-guarded episodes as assisted.",{"version":1536,"date":1529,"sections":1537},"18.0.28",[1538],{"name":121,"changes":1539},[1540,1541],"Add opt-in, bounded Crystal stall rescue for the authenticated Mystery Egg return, reusing the audited map fields and verifying the shared Pokémon Center second-floor warp before assistance.","Keep battle, dialogue, item use and Egg delivery under native owners. Rescue actions remain controller-owned and excluded from policy navigation credit; preserved weights and Adam state require a managed compatible-source handoff before live use.",{"version":1543,"date":1529,"sections":1544},"18.0.27",[1545],{"name":121,"changes":1546},[1547,1548],"Route low-HP Red recovery around the displaced Pewter Gym Junior Trainer through the verified east aisle, preventing repeated attempts to enter the trainer's tile.","Verify the trainer's native approach, battle victory, recovery-owned gym exit and Pokémon Center heal from an unmodified retained state. Preserve learned weights, Adam state, checkpoint lineage and qualification authority; live activation remains subject to the managed source handoff.",{"version":1550,"date":1529,"sections":1551},"18.0.26",[1552],{"name":121,"changes":1553},[1554,1555,1556,1557,1558],"Distinguish each campaign's remaining storage allocation, physical disk free space and the host's unreserved buffer. Forecast allocation use and unreserved-buffer depletion from their own measured changes.","Display the separate storage quantities without treating legacy headroom as the space remaining for training.","Publish selected campaign observations through a separate telemetry identity, preserving producer timestamps and staleness without competing with trainer heartbeats.","Add reviewed deduplication of verified immutable files in inactive engineering runs, preserving component paths, checkpoint contents and diagnostic evidence.","Preserve learner weights, Adam state, qualification policy, storage allocations and the host safety reserve. Runtime activation uses the managed compatible-source handoff.",{"version":1560,"date":1529,"sections":1561},"18.0.25",[1562],{"name":121,"changes":1563},[1564,1565,1566],"Add reviewed, fenced reconciliation of a safety-stopped campaign to its latest complete learner\u002FAdam\u002Fmemory bundle before a compatible source update.","Preserve safety incidents and charged work; record unsaved progress as lost instead of launching an old runtime merely to make it pausable.","Retain the Crystal return-to-Elm, Red capability fallback and learning-credit changes from 18.0.24.",{"version":1568,"date":1529,"sections":1569},"18.0.24",[1570],{"name":121,"changes":1571},[1572,1573,1574,1575],"Extend Crystal's audited post-Egg return guidance and legal recovery coverage toward New Bark and Elm, preserving native story events and action ownership.","Separate Red's finite preparation release from surviving-party fighting capability, with bounded fallbacks for depleted PP or missing damaging moves.","Record quest-waypoint learning credit by game, starter and action owner so Crystal return phases and uneven Red frontier exposure can be measured.","Preserve learned weights, Adam state, model lineage and campaign integrity holds. This source change does not install, deploy or activate a new runtime.",{"version":1577,"date":1529,"sections":1578},"18.0.23",[1579],{"name":121,"changes":1580},[1581,1582,1583,1584],"Show recorded Red gym visits and badge acquisitions in the Sankey even when an attempt later returns to town and stalls. Keep unknown badge identity and pre-existing badges distinct from new victories.","Continue badge-backed Pewter guidance through interior exits toward Route 3, without crediting missed earlier waypoints or changing the +6,000 badge bonus.","Finish a verified Brock victory's nondecision badge dialogue within its initiating environment step, preserving actual action ownership and native victory evidence before recovery resumes.","Preserve learned weights, optimizer state and checkpoint lineage. Source release, website deployment and managed training activation are separate stages.",{"version":1586,"date":1587,"sections":1588},"18.0.22","2026-09-21",[1589],{"name":121,"changes":1590},[1591,1592,1593],"Expand detailed run Pokémon cards with recorded moves, move types, native current\u002Fmax PP, DVs, condition, held items, capture ball and throw count, level progression, observed evolutions, and faint context.","Preserve each individual's latest recorded profile after leaving the party. Keep unknown historical values and unproven finishing moves explicit, and use generation-specific Red\u002FCrystal data without synthesizing natures.","Record presentation telemetry separately from learner inputs. This source release does not install, deploy, or activate a training runtime; campaign activation uses an immutable capsule and managed handoff.",{"version":1595,"date":1587,"sections":1596},"18.0.21",[1597],{"name":121,"changes":1598},[1599,1600],"Add explicit managed recovery into a new campaign from a selected historical checkpoint. Preserve the previous campaign's sealed learner and history, fence its authority, and transfer scheduler selection only after verified admission.","Bind recovery to the exact predecessor, selected policy, optimizer and declared auxiliary memory. Retain archived storage accounting and the single-trainer constraint; ordinary continuation never chooses an older checkpoint automatically.",{"version":1602,"date":1587,"sections":1603},"18.0.20",[1604],{"name":121,"changes":1605},[1606,1607],"Reject Crystal campaign starter resets whose state-to-waypoint binding is missing or differs from the canonical opening. This catches the migrated path mismatch that skipped the opening quest and incorrectly targeted returning the Mystery Egg.","Repair live configuration through an immutable runtime and managed learner-preserving handoff. Preserve policy weights, Adam, navigation memory and historical outcomes; keep source, installation and activation evidence separate.",{"version":1609,"date":1587,"sections":1610},"18.0.19",[1611],{"name":121,"changes":1612},[1613,1614],"Fix the managed OpenClaw launcher installer by importing the immutable-metadata projection it uses. Add a regression that installs and validates a sealed launcher from admitted campaign authority.","Preserve the already verified 18.0.18 learner capsules and their exact performance evidence. This installer correction changes no PPO, environment, optimizer, qualification, or model state; source\u002Finstaller release and active learner runtime are reported separately.",{"version":1616,"date":1587,"sections":1617},"18.0.18",[1618],{"name":121,"changes":1619},[1620,1621],"Separate configured legal tactical-risk telemetry from genuine native rule and action-execution violations in the campaign runtime guard. A controller's `forced_unsafe_attack` observation alone no longer falsely holds the learner when the resolved rules permit that action.","Preserve tactical-controller behavior, PPO\u002FAdam state, model lineage, qualification rules and native legality enforcement. Installation and activation require a new immutable runtime capsule and managed handoff; this source release does not modify an existing 18.0.17 capsule.",{"version":1623,"date":1587,"sections":1624},"18.0.17",[1625],{"name":121,"changes":1626},[1627,1628,1629],"Add opt-in renewable Red\u002FCrystal campaigns with persistent PPO\u002FAdam learning, complete post-update snapshots, background evaluations and separate protected promotion.","Preserve legacy holds and measured failures through explicit managed migration; add fenced recovery, immutable runtime epochs, bounded host resources and truthful campaign status.","Integrate the 18.0.16 external log storage release. Installation, scheduler handoff and optimizer activation are recorded separately in rollout evidence.",{"version":1631,"date":1587,"sections":1632},"18.0.16",[1633],{"name":121,"changes":1634},[1635,1636,1637,1638],"Add bounded, compressed training log segments and separately compressed battle evidence, with complete-history readers.","Prefer verified external storage for bulk log payloads, fall back locally on device failure, and retain upload identities and receipts.","Back off repeated uploader failures, summarize console errors, and keep SB3 metric collection with compact console rendering.","Reserve local checkpoint\u002Fnavigation capacity during active training and preserve an interrupted managed hold on storage pressure.",{"version":1640,"date":1587,"sections":1641},"18.0.15",[1642],{"name":121,"changes":1643},[1644,1645],"Recover reviewed Red post-checkpoint disk-full navigation finalization with separate immutable archive receipts, preserving the original failed manifest, terminal policy and Adam state, journal lineage, and existing qualification\u002Fone-use launch gates.","Hold new Python-supervised work when archive capacity is insufficient; resume storage-only holds after capacity returns and include underlying errors in Slack pause alerts.",{"version":1647,"date":1587,"sections":1648},"18.0.14",[1649],{"name":121,"changes":1650},[1651],"Recognize an exact native trainer during startup even when its claim predates launch by several minutes and health still names the previous batch. Bound startup by real process age, preserve failure gates, and retain reviewed false-alert evidence without suppressing later alerts.",{"version":1653,"date":1587,"sections":1654},"18.0.13",[1655],{"name":121,"changes":1656},[1657,1658],"Add an immutable Python supervisor for exact managed continuation, cheap active-run checks, durable pause evidence, and deduplicated Slack alerts.","Replace routine LLM continuation monitoring with native command jobs; preserve all qualification gates and bounded training budgets.",{"version":1660,"date":1587,"sections":1661},"18.0.12",[1662],{"name":121,"changes":1663},[1664,1665],"Repair Crystal frozen qualification readiness for normalized launch metadata.","Add exact, reviewed zero-result preflight recovery preserving the pinned reference and managed qualification gate.",{"version":1667,"date":1587,"sections":1668},"18.0.11",[1669],{"name":121,"changes":1670},[1671,1672,1673],"Use a shared frozen opening qualification policy for Red and Crystal, with starter and seed-group success floors and a one-case count margin against a fixed reference. Ordinary legal losses no longer count as safety violations; illegal behavior and incomplete or corrupt evidence still block continuation.","Qualify Crystal's three natural starter states through native Mr. Pokémon house entry, pin the reference, and bind every managed launch to its exact passed receipt. Preserve existing behavioral holds and one-use admissions.","Add an explicit managed policy upgrade for Red's held learned terminal and Crystal's completed trials, preserving weights, optimizer state and journals. Historical failures remain failures and never acquire current launch authority.",{"version":1675,"date":1676,"sections":1677},"18.0.10","2026-09-20",[1678],{"name":121,"changes":1679},[1680,1681,1682],"Give Red's waypoint-17 opening navigation a collision-audited Route 1 field, including the sideways detour around ledges. Preserve policy-owned movement, semantic milestones and high-water progress accounting; cover the previously untested waypoint-16-to-18 handoff with native cartridge replays.","Let an explicit source-remediation review requalify the same rejected Red checkpoint and frozen journal after this exact navigation fix. Preserve failed evidence, learned weights, Adam state and recovery history; require fresh all-starter qualification before one managed continuation.","Give the existing Red owner up to one hour for complete nested admission proofs, including silent validation, so its former timeout cannot interrupt a managed launch after its one-use claim has been consumed.",{"version":1684,"date":1676,"sections":1685},"18.0.9",[1686],{"name":121,"changes":1687},[1688],"Add an explicit one-use Red reproducibility review after a learned successor fails frozen qualification. Restore only its immediately preceding qualified learned policy, Adam state and matching immutable pre-batch trainer journal; preserve rejected weights, original holds and admission history. Require a fresh current-source all-starter qualification before the managed launch, and prevent another recovery from the same learned parent.",{"version":1690,"date":1676,"sections":1691},"18.0.8",[1692],{"name":31,"changes":1693},[1694,1695,1696,1697,1698,1699],"Publish bounded per-episode diagnostics and version 18 battle advice for the detailed run page, preserving unknown values, censored outcomes and exact native Pokémon identities.","Keep raw terminal traces, item windows and local filesystem references out of the public telemetry projection; report size-related omissions explicitly.","Keep the run modal concise and link to a dedicated page with recorded stats, Pokémon details and advisor evidence.","This is a source release only. Installation, dashboard deployment, historical telemetry republication, qualification and managed training activation remain separate operations.","Allow an explicit Crystal guardian recovery review after a sealed zero-step environment permission failure of an already trained advisor continuation. Preserve the exact completed policy, Adam state, learned trainer journal, original behavioral hold, failed execution and consumed claim. A new bounded admission is single-use and revalidates the archived failure before launch.","Keep legacy advisor migration recovery unchanged. Runtime\u002Fsource\u002Fconfiguration or ancestry mismatches, nonzero work, active children and duplicate retries still hold; training rules, rewards, controllers and PPO settings are unchanged.",{"version":1701,"date":1676,"sections":1702},"18.0.7",[1703],{"name":121,"changes":1704},[1705,1706],"Admit reviewed Red and Crystal successor trials from the exact learned advisor terminal checkpoint, preserving optimizer state, feature contracts and trainer memory. Each bounded batch retains its original hold and receives a new one-use guardian review; Red still requires exact opening qualification.","Distinguish continued advisor checkpoint provenance from the initial zero-input migration. Refresh source, configuration and runtime bindings on installation.",{"version":1708,"date":1676,"sections":1709},"18.0.5",[1710],{"name":121,"changes":1711},[1712,1713],"Add an explicitly reviewed Red fallback admission after failed advisor qualification, preserving rejected weights, optimizer ancestry and fresh qualification requirements.","Retain a bounded terminal overworld trace for semantic evaluations so navigation stalls can be reviewed without changing gameplay or scoring.",{"version":1715,"date":1676,"sections":1716},"18.0.4",[1717],{"name":121,"changes":1718},[1719,1720,1721],"Verify admitted Crystal advisor parents during worker readiness, preserving native lineage checks and the exact migration provenance.","Check every worker’s readiness before consuming a guardian launch claim.","Preserve the sealed zero-step environment failure and attempt history when explicitly reviewing one new advisor admission; keep Red’s active 18.0.1 qualification unchanged.",{"version":1723,"date":1676,"sections":1724},"18.0.3",[1725],{"name":121,"changes":1726},[1727,1728,1729],"Make Crystal guardian path setup idempotent, preserving identical package discovery when the file entry point is imported again by trainer preflight.","Preserve both diagnosed pre-execution failures and require a fresh reviewed admission for the known 18.0.2 duplicate-metadata identity failure.","Keep Red’s running 18.0.1 qualification unchanged.",{"version":1731,"date":1676,"sections":1732},"18.0.2",[1733],{"name":121,"changes":1734},[1735,1736,1737],"Fix the Crystal guardian script entry point so its trainer can verify the exact managed child claim.","Preserve and explicitly review a proven pre-execution child import failure before preparing one new admission; never replay the consumed claim.","Keep Red’s already running 18.0.1 frozen qualification and its source unchanged during the Crystal repair.",{"version":1739,"date":1676,"sections":1740},"18.0.1",[1741],{"name":121,"changes":1742},[1743,1744,1745],"Admit verified battle-advisor checkpoint expansions through Red and Crystal's managed guardians while preserving original review holds, native parent lineage and one-use trial claims.","Seal exact preparation base configurations and reject altered recipes, migrated weights, optimizer history, source, configuration or seed policy before activation.","Preserve cumulative lineage for admitted migrated checkpoints and reject unreviewed recipe launches or fresh-weight fallbacks.",{"version":1747,"date":1676,"sections":1748},"18.0.0",[1749],{"name":121,"changes":1750},[1751,1752,1753,1754,1755],"Add opt-in Red and Crystal capture\u002Fdamage advice with pinned generation-specific mechanics and explicit unsupported cases.","Expose native DVs, stat experience, actual stats and level experience for identified party members; natures remain unsupported in these generations.","Add observed-only persistent trainer scouting, frozen evaluation snapshots, party planning and separately attributed reward experiments.","Prepare advice feature migration and bounded matched experiments while preserving existing managed activation gates.","Designate the battle-learning source as version 18.0.0 while preserving separate Red and Crystal checkpoint lineages. Prior 17.0.45 receipts remain historical; managed queueing and current-source qualification are pending.",{"version":1757,"date":1758,"sections":1759},"17.0.44","2026-09-19",[1760],{"name":121,"changes":1761},[1762],"Repair artifact uploads when identical bytes and filenames occur in different execution directories. Retry the ownership conflict once with an artifact-ID scoped object key, preserving existing mappings, retirement records, and fail-closed project\u002Frun ownership checks.",{"version":1764,"date":1758,"sections":1765},"17.0.43",[1766],{"name":31,"changes":1767},[1768,1769,1770,1771],"Merge the current Red retention and Crystal recovery, reviewed-trial and telemetry changes into one source release.","Preserve cloud trace identities across backfill retries after partial ingestion failures.","Accept legacy dual telemetry only when validating sealed Crystal predecessors; require cloud-primary telemetry for new launches.","Preserve both games' immutable evidence and use managed qualification and reviewed-trial resumption.",{"version":1773,"date":1758,"sections":1774},"17.0.42",[1775],{"name":1198,"changes":1776},[1777,1778],"Validate sealed Crystal predecessors against their historical dual-storage contract so a completed pre-migration trial can receive an explicit managed successor review. New launches still require isolated cloud-primary storage.","Preserve exact checkpoints, optimizer lineage, immutable predecessor evidence, original behavioral holds, and one-use reviewed trial claims.",{"version":1780,"date":1758,"sections":1781},"17.0.41",[1782],{"name":31,"changes":1783},[1784,1785,1786],"Name new local step logs by numeric release and execution, and read legacy plus versioned history without losing archived evidence.","Include the numeric release in new cloud trace filenames and manifests; preserve historical provenance during backfill.","Give legacy compressed logs a named archive path with a canonical compatibility link, retaining exact bytes and review evidence.",{"version":1788,"date":1758,"sections":1789},"17.0.40",[1790],{"name":31,"changes":1791},[1784,1785,1786],{"version":1793,"date":1758,"sections":1794},"17.0.39",[1795],{"name":31,"changes":1796},[1797,1798,1799],"Stage Crystal cloud-primary trace storage using the existing bounded, execution-scoped gzip uploader and emergency trace fallback. Preserve all 17.0.35–17.0.37 Crystal policy, recovery, cohort and reviewed-trial safeguards.","Preserve historical canonical step logs as verified gzip archives, make local analysis archive-aware, and require explicit restoration before local tracing or provenance-sensitive backfill can reopen those logs.","Keep this release staged until the active reviewed trial finishes and seals. Installation, historical compression and optimizer activation are separate managed operations; no active trainer or trial identity changes during staging.",{"version":1801,"date":1758,"sections":1802},"17.0.38",[1803],{"name":1804,"changes":1805},"Added",[1806,1807],"Read verified archived CSV history without hiding it, refuse unsafe writer reopen or partial backfill, and provide guarded lossless archival of closed step logs.","Add opt-in, bounded overworld decision evidence for investigating pre-Parcel navigation failures without changing policy decisions or random-number consumption.",{"version":1809,"date":1758,"sections":1810},"17.0.37",[1811],{"name":1198,"changes":1812},[1813,1814],"Extend bounded low-HP recovery along the audited post-Egg return route, retaining native healing and battle-rule guards.","Persist actor\u002Fvalue sample counts and controller-mask fractions in optimizer history.",{"version":1816,"date":1758,"sections":1817},"17.0.36",[1818],{"name":121,"changes":1819},[1820,1821],"Add opt-in, policy-owned Crystal return guidance inside Mr. Pokémon's house and south along Route 30 after receiving the Mystery Egg. Preserve real map arrivals and story evidence, and retain once-only progress across return waypoint advances. Keep the opening projection scope unchanged.","Preserve the completed 17.0.35 trial and native phase\u002Fstarter\u002Fowner audit; authenticate both retained house-stall replays and pin house geometry against the ROM. A successor remains a separate, bounded guardian-reviewed trial.",{"version":1823,"date":1758,"sections":1824},"17.0.35",[1825],{"name":121,"changes":1826},[1827,1828],"Extend opt-in Crystal stall rescue through the exact pre-Egg Route 30 house objective using the existing audited field and unchanged rescue budgets. Keep the real doorway transition controller-owned and release before story dialogue; assisted episodes remain ineligible for promotion.","Preserve the completed 17.0.34 trial, full native phase\u002Fstarter\u002Fowner audit, optimizer lineage and original behavioral hold for managed successor review.",{"version":1830,"date":1831,"sections":1832},"17.0.34","2026-09-18",[1833],{"name":121,"changes":1834},[1835,1836],"Allow an explicitly reviewed successor after a completed Crystal trial. Bind the exact prior child, source, configuration, start record, sealed optimizer budget and terminal checkpoint; preserve the original hold and consume each new trial once. Never repeat a finished trial automatically.","Extend existing bounded, verified menu cancellation to Cherrygrove's Gym Speech House and Evolution Speech House during the exact pre-Egg journey. Preserve protected dialogue, UI proof, dwell and intervention limits, policy interfaces and compatible optimizer lineage. This corrects missing menu coverage; it does not establish improved policy navigation.",{"version":1838,"date":1831,"sections":1839},"17.0.33",[1840],{"name":121,"changes":1841},[1842,1843],"Accept archived JSON audit receipts under `docs\u002Fsuperpowers\u002Fevidence\u002F` as documentation during a managed Red release handoff. Keep source hashes, clean-install checks, ancestry and every intermediate commit checked; scripts, runtime configuration and JSON elsewhere still block same-release replacement.","Preserve the exact interrupted weights and optimizer lineage. Requalify the candidate and fallback under the installed release before managed continuation.",{"version":1845,"date":1831,"sections":1846},"17.0.32",[1847],{"name":121,"changes":1848},[1849],"Accept an explicitly bound, sealed same-Crystal checkpoint in readiness; preserve failed zero-step reviewed launches and require explicit failure evidence to prepare a replacement trial.",{"version":1851,"date":1831,"sections":1852},"17.0.31",[1853],{"name":121,"changes":1854},[1855],"Report Crystal UI learning phases precisely and support one immutable reviewed 500k trial while preserving the behavioral hold and starting optimizer weights.",{"version":1857,"date":1831,"sections":1858},"17.0.30",[1859],{"name":121,"changes":1860},[1861],"Preserve short-batch optimizer candidates when completed starter coverage is insufficient; require exact frozen retention and hold failed trials instead of silently repeating rollback runs.",{"version":1863,"date":1831,"sections":1864},"17.0.29",[1865],{"name":121,"changes":1866},[1867,1868],"Qualify the actual Viridian-to-Route2 map transition after Parcel delivery, independently of a lagging route pointer. Preserve observed before\u002Fafter map proof and all starter, living-roster, order and native-safety checks.","Preserve the incomplete17.0.27 panel as diagnostic evidence of the remaining scorer gap; require new frozen qualification under the corrected source.",{"version":1870,"date":1831,"sections":1871},"17.0.28",[1872],{"name":121,"changes":1873},[1874,1875,1876,1877],"Limit new managed Crystal continuation batches to 500,000 optimizer steps, preserving any unfinished sealed lineage target and the existing 250,000-step checkpoint interval.","Persist a managed behavioral review hold for repeated pre-house stalls across all three starters; bind bounded index evidence and terminal checkpoint identity. Missing evidence holds, and legacy resumes retain their sealed execution budget.","Bound normal Crystal inference-only drain to 120 seconds. Preserve optimizer counters, report drain decisions separately, and censor unfinished episodes without recording fabricated completions, stalls, or legal successes.","Audit current Crystal navigation stalls with exact retained-state replays and a frozen Route 30 checkpoint panel; keep navigation\u002Freward behavior unchanged because those observations do not establish another controller defect.",{"version":1879,"date":1831,"sections":1880},"17.0.27",[1881],{"name":121,"changes":1882},[1883,1884,1885,1886],"Accept verified Route 2 waypoint synchronization as ordered opening evidence, retain exact proof, and test the transition against real cartridge gameplay.","Match primary retention rules to training and report strict Nuzlocke challenge evaluation separately. Qualify fallback and promotion weights with held-out seeds; preserve immutable evidence and report per-starter denominators.","Bound new ordinary Red learning batches to 500,000 requested optimizer steps with 250,000-step checkpoints and a 120-second completion drain. Preserve sealed legacy budgets; distinguish optimizer progress and inference draining.","Record immutable trial inputs and reject unmatched or multi-change claims of controlled comparison. Audit existing learning credit by phase, starter and action owner, with aligned optimizer metrics and explicit evidence coverage.",{"version":1888,"date":1831,"sections":1889},"17.0.26",[1890],{"name":121,"changes":1891},[1892,1893,1894],"Require exact saved Red checkpoints to retain Parcel delivery and the Viridian north exit on a frozen starter-balanced panel before ordinary continuation, recovery, fallback, or launch. Bind and revalidate the underlying evidence; missing, stale, incomplete, or failed evidence holds training.","Stop sustained early collapse at a completed optimizer update, then use a bounded episode drain and immutable checkpoint. Keep unfinished cutoffs separate from completed policy failures, and qualify fallback weights too.","Enforce sealed evaluation evidence inside trainer startup and document the required release and continuation process. Preserve compatible model weights, optimizer state, and the existing full-run lineage.",{"version":1896,"date":1897,"sections":1898},"17.0.25","2026-09-17",[1899],{"name":121,"changes":1900},[1901,1902,1903],"Preserve Crystal's opening-route reward history after Route 30 arrival, so returning through previously credited tiles cannot earn duplicate reward or renew the meaningful-progress timeout. Keep fresh northbound progress rewarded.","Add audited, bounded exit assistance for Cherrygrove's Pokémon Center floors while the exact pre-Egg objective is pending. Verify the shared upstairs destination and preserve existing intervention and action limits.","Keep learned weights, optimizer state, observation\u002Faction interfaces and the `pokemon-crystal-starters-v1` lineage compatible. See [training notes](docs\u002Ftraining-17.0.25.md) for evidence and release stages.",{"version":1905,"date":1897,"sections":1906},"17.0.24",[1907],{"name":121,"changes":1908},[1909,1910,1911],"Recognize verified Crystal overworld object-update frames so ordinary field movement does not supply a false menu flag to the policy.","Preserve protected menu\u002Fdialogue ownership, recovery limits, Nuzlocke rules, policy interfaces and compatible learned weights through managed activation.","Add a retained-state regression for the observed Center 2F classification gap.",{"version":1913,"date":1897,"sections":1914},"17.0.23",[1915],{"name":121,"changes":1916},[1917,1918],"Restore Red policy guidance on nine surveyed northern Route 2 tiles outside the existing corridor's snap radius. Rejoin the verified gate column while preserving semantic progress and shared high-water credit against repeat boundary rewards.","Retain the two exact 17.0.20 Route 2 failures with recorded-input replays and ordinary-input escape checks from all eight uncovered recorded positions.",{"version":1920,"date":1897,"sections":1921},"17.0.22",[1922],{"name":121,"changes":1923},[1924,1925,1926],"Allow bounded cancellation of verified incidental menus in Cherrygrove Pokémon Center 1F while the shipped pre-Egg objectives remain pending. Preserve healing priority, protected dialogue, action budgets and actor masking.","Retain compact diagnostic history preceding the failure replay checkpoint; keep exact replay inputs separate and disclose context truncation.","Preserve a digest-verified Crystal Center state and recorded inputs, with regressions for menu closure, protected UI and compatible policy interfaces.",{"version":1928,"date":1897,"sections":1929},"17.0.21",[1930],{"name":121,"changes":1931},[1932,1933,1934],"Keep bounded Crystal opening-route and incidental-menu recovery available after Route 30 arrival when the exact Mr. Pokémon house objective is pending. Restore guidance out of early interiors without rewinding semantic progress.","Preserve existing action\u002Fintervention budgets, protected UI and story guards, controller actor masking, Nuzlocke rules and compatible checkpoint lineage.","Retain three digest-verified release-17.0.18 stall states and recorded-input replays covering New Bark backtracking and a home-interior guidance gap.",{"version":1936,"date":1937,"sections":1938},"17.0.20","2026-09-16",[1939],{"name":121,"changes":1940},[1941,1942],"Guide early Red navigation around surveyed Pallet and southern Route 1 obstacles using ordinary policy inputs and local route targets. Preserve semantic progress, checkpoint lineage and existing battle\u002Fcontroller rules.","Include 17.0.19's observation-target diagnostics in the managed Red release.",{"version":1944,"date":1937,"sections":1945},"17.0.19",[1946],{"name":121,"changes":1947},[1948,1949],"Record the actual `wp_distance` observation target before and after each action in failure capsules, including route identity and raw target distance. Keep off-phase reward potential separate and preserve missing guidance as unknown, without extra navigation lookups or changes to policy inputs.","Retain three digest-verified early Red stall captures with recorded-input replays and ordinary-input detour checks. The captures reproduce wandering in Pallet Town and Route 1; no reward or controller changes are inferred from these limited diagnostics.",{"version":1951,"date":1937,"sections":1952},"17.0.18",[1953],{"name":121,"changes":1954},[1955,1956,1957],"Extend pre-Egg Crystal field Potion use to Route 30 and Cherrygrove, with a bounded, audited retreat to Cherrygrove's nurse when field healing is unavailable. Restore living-party HP, status and PP and finish dialogue before returning outside. Preserve rule guards, controller attribution and reward suppression.","Guide the pending Mr. Pokémon objective along collision\u002Fevent-audited Route 30 tiles, and resume northbound guidance after healing without rewinding progress. Keep later story phases, custom objectives and unsupported terrain excluded.","Retain ordinary-input regressions for natural injury, Potion and nurse healing, complete south\u002Fnorth navigation and the actual house warp. Learned weights, checkpoint lineage and policy interfaces remain compatible.",{"version":1959,"date":1937,"sections":1960},"17.0.17",[1961],{"name":121,"changes":1962},[1963,1964],"Reuse Crystal's audited Cherrygrove northbound field in bounded stall rescue while Route 30 is the pending frontier, and release control at the real map connection. Retain existing action, intervention and cooldown limits.","Preserve battle, healing and protected-UI ownership, unsupported-interior guards, reward suppression and actor masking. Assisted attempts remain ineligible for promotion; learned weights and policy interfaces are unchanged.",{"version":1966,"date":1937,"sections":1967},"17.0.16",[1968],{"name":121,"changes":1969},[1970],"Restore audited southern Route 2 and Forest-gate guidance when preparation reaches them with a missed post-Parcel Viridian waypoint. Preserve pending semantic waypoints and bound repeated navigation rewards.",{"version":1972,"date":1937,"sections":1973},"17.0.15",[1974],{"name":121,"changes":1975},[1976,1977,1978],"Restore a safe damage-capable fighter once per battle when a released preparation phase leaves an unusable lead, while preserving policy ownership of subsequent attacks.","Extend Red's ordered route from Pewter arrival to the Gym and the first real badge. Restore local guidance for missed Forest anchors, the Museum, and the Center using ordinary-input collision audits.","Add bounded cohort comparisons with equal starter weighting, valid-ending deduplication, and the original waypoint scale capped at 70 so new route anchors cannot inflate the comparison.",{"version":1980,"date":1937,"sections":1981},"17.0.14",[1982],{"name":121,"changes":1983},[1984,1985],"Require an observed positive enemy HP baseline before crediting damage. Crystal battle setup can briefly report zero HP with an initialized maximum; those samples no longer award false damage or consume the wild reward cap.","Preserve observed damage, knockout and explicit battle-result credit, and reset the HP baseline for new battles, episodes and changed enemy species. Keep policy interfaces, learned weights and checkpoint lineage compatible.",{"version":1987,"date":1937,"sections":1988},"17.0.13",[1989],{"name":121,"changes":1990},[1991,1992],"Keep existing Crystal Potion, bounded wild escape, and audited Elm healing eligible after a Route 30 visit when the policy returns to Route 29, New Bark, or Elm's Lab before receiving the Mystery Egg.","Preserve existing map\u002Fstory guards, guidance, rescue budgets, policy interfaces and checkpoint lineage. Installation and managed activation are separate from this source release.",{"version":1994,"date":1937,"sections":1995},"17.0.12",[1996],{"name":121,"changes":1997},[1998,1999,2000],"Fix Red Center recovery for parties with full HP but depleted move PP. Check known move PP when deciding whether to heal and when verifying fresh or resumed nurse dialogue, so low-PP recovery cannot loop as already healthy.","Retain both original Viridian Center failures as diagnostic fixtures. Their ordinary nurse inputs restore PP, preserve the living roster and Nuzlocke ledgers, and return subsequent movement to the policy.","Preserve recovery thresholds, controller ownership, compatible learned weights and `nuzlocke-full-run-recurrent-v3`. Source preparation, installation and managed next-batch activation remain separate stages.",{"version":2002,"date":1937,"sections":2003},"17.0.11",[2004],{"name":121,"changes":2005},[2006,2007,2008],"Restore Red's north Forest gate guidance after a missed post-Parcel anchor. Use the retained failure's verified western and center aisle, keeping unknown tiles outside the projection and preserving the real UP doorway action.","Join the gate's verified northern Route 2 arrival column to the existing Pewter corridor with continuous, once-only route progress shaping.","Keep policy ownership, semantic prerequisites, story and Nuzlocke ledgers, compatible learned weights and `nuzlocke-full-run-recurrent-v3` unchanged. Record source preparation separately from installation and live activation.",{"version":2010,"date":2011,"sections":2012},"17.0.10","2026-09-15",[2013],{"name":121,"changes":2014},[2015,2016],"Add phase-scoped, collision-audited reward guidance from Cherrygrove to Route 30, covering the missing local route after the first Cherrygrove visit. Preserve the semantic Route 30 arrival milestone and once-only movement rewards, with ordinary-input ROM replays and no added controller ownership.","Keep learned weights, model lineage, policy interfaces and existing rescue limits compatible; installation and managed activation are separate stages.",{"version":2018,"date":2011,"sections":2019},"17.0.9",[2020],{"name":121,"changes":2021},[2022,2023,2024],"Restore Red's existing northern Route 2 guidance when a missed post-Parcel anchor remains pending. Preserve the semantic pointer, story prerequisites, policy ownership and once-only navigation progress rewards.","Add the retained western edge's verified down\u002Fright rejoin to the existing Pewter corridor. Keep unknown tiles outside the route.","Preserve compatible learned weights and `nuzlocke-full-run-recurrent-v3`. Source release, installation and managed live activation remain separate.",{"version":2026,"date":2011,"sections":2027},"17.0.8",[2028],{"name":121,"changes":2029},[2030,2031],"Keep audited Crystal opening-map guidance, bounded route\u002Fmenu rescue and low-HP recovery available after the first Cherrygrove visit while Route 30 remains pending. Backtracking no longer silently disables those protections.","Preserve controller budgets, quest\u002Fmap guards, promotion exclusions, learned weights, optimizer state and the existing Crystal lineage.",{"version":2033,"date":2011,"sections":2034},"17.0.7",[2035],{"name":121,"changes":2036},[2037,2038],"Resume a suspended Crystal opening-route rescue after healing without spending another intervention. Preserve its original action count and elapsed deadline, so frequent recovery handoffs cannot discard most of the navigation budget or create unbounded assistance.","Record whether an assisted episode's rescue is suspended. Preserve the policy contract, learned weights, promotion exclusions and lineage.",{"version":2040,"date":2011,"sections":2041},"17.0.6",[2042],{"name":121,"changes":2043},[2044,2045],"Respect explicit promotion exclusions independently of static-start settings in live milestones, completed and censored episodes, historical CSV loading, and curriculum row\u002Fsidecar gates. Crystal rescue progress remains visible without counting as independent policy promotion evidence.","Preserve the 17.0.5 rescue behavior, learned weights, optimizer and model lineage. See `docs\u002Ftraining-17.0.6.md` for verification and activation.",{"version":2047,"date":2011,"sections":2048},"17.0.5",[2049],{"name":121,"changes":2050},[2051,2052],"Add bounded Crystal opening-route rescue after 2,048 decisions without meaningful progress. Follow pinned-ROM floor routes through New Bark, early house detours, and Route 29 using guarded ordinary inputs. Preserve battle, dialogue and item ownership; cap interventions and return control to policy.","Mask rescue actions from actor training and suppress their navigation rewards. Keep assisted episodes out of checkpoint promotion and retain rescue details in diagnostic traces. Preserve the existing checkpoint and observation lineage.",{"version":2054,"date":2011,"sections":2055},"17.0.4",[2056],{"name":121,"changes":2057},[2058,2059,2060],"Separate Red controller synchronization failures from evidence of worse learned gameplay. Quarantine affected batches from promotion and recover the exact verified pre-batch continuation instead of discarding recent learning in favor of a much older accepted-best checkpoint.","Hold training when recovery evidence is missing or changed, or when technical failures recur. Keep recovery accounting persistent and idempotent; retain strict promotion and non-technical regression checks.","Preserve saved checkpoints and the `nuzlocke-full-run-recurrent-v3` lineage. Installation and live recovery use the managed version handoff.",{"version":2062,"date":2011,"sections":2063},"17.0.3",[2064],{"name":121,"changes":2065},[2066,2067,2068],"Preserve Crystal's proven incidental-menu dwell across up to 32 unowned transition or overworld steps at the same position. Unknown states never receive automatic input or add dwell; story\u002Fitem transactions, movement, long gaps and other controllers reset it. Keep cancellation budgets, cooldown, rule guards, actor-loss masking and the existing observation contract.","Allow same-release managed Red handoffs across verified documentation-only descendant commits when the Python source hash still matches. Inspect every intervening commit and reject non-documentation changes or missing history.","Preserve Red and Crystal learned checkpoint lineages; activate through the managed handoff after installation and validation.",{"version":2070,"date":2071,"sections":2072},"17.0.2","2026-09-14",[2073],{"name":1198,"changes":2074},[2075,2076,2077],"Decode Crystal field UI from validated live ROM call chains instead of stale menu flags; preserve the calibrated state and proof in failure capsules.","Escape idle, verified incidental menus during the New Bark\u002FRoute 29 opening with bounded guarded B inputs. Preserve health-recovery priority, exclude unknown UI, story dialogue and item transactions, and mask controller actions from policy learning and navigation reward credit.","Preserve the Crystal checkpoint lineage, 23 policy features and eight actions.",{"version":2079,"date":2071,"sections":2080},"17.0.1",[2081],{"name":1198,"changes":2082},[2083,2084,2085,2086,2087],"Complete the verified north Forest doorway crossing for a ready Red team, retaining bounded handoff, fade, movement, and Nuzlocke guards.","Bound preparation-owned transition waits even when readiness pauses the normal preparation budget.","Guide stale Forest\u002Fgate waypoints through northern Route 2's replayed Pewter corridor instead of the disconnected southern gate, without skipping semantic milestones or repaying backtracking.","Preserve before\u002Fafter menu and battle state plus preparation decisions in bounded failure capsules for both games.","Keep both learned policy lineages and action\u002Fobservation contracts unchanged.",{"version":2089,"date":2071,"sections":2090},"17.0.0",[2091],{"name":121,"changes":2092},[2093,2094,2095],"Replace the Red lifetime navigation-map limit with indexed persistent storage, preserving legacy observations and contextual safety evidence.","Add independently calibrated Crystal map learning and same-game sharing across starters, with guarded learned recovery to verified facilities.","Establish a 17.0.0 comparison boundary while preserving compatible Red and Crystal policy lineages and recording source, installation, and activation separately.",{"version":2097,"date":2071,"sections":2098},"16.8.32",[2099],{"name":121,"changes":2100},[2101,2102],"Add collision-audited local reward guidance from New Bark through Route 29 to Cherrygrove, including detours observed in the running starter cohort. Episode-best route-distance rewards teach necessary bends without paying extra for repeated backtracking. Actions, observations and checkpoint lineage stay compatible.","Correct Route 29’s west exit anchor to (0,6), covering the real connection at rows 6–7, and tighten both opening transition radii to one tile. Actual map crossings still own semantic waypoint completion; healing remains controller-masked.",{"version":2104,"date":2071,"sections":2105},"16.8.31",[2106],{"name":1198,"changes":2107},[2108,2109,2110],"Settle pending Crystal policy movement before claiming the early healing corridor, re-read the actual tile, and yield unsupported arrivals without ending an otherwise valid episode. Keep the wait under controller ownership.","Specify westward opening map connections so returning from New Bark to Route 29 receives its existing one-time off-phase exit reward.","Preserve checkpoint lineage, learned tensors, observation\u002Faction shapes and Nuzlocke enforcement. See [release evidence](docs\u002Ftraining-16.8.31.md).",{"version":2112,"date":2113,"sections":2114},"16.8.30","2026-09-13",[2115],{"name":1198,"changes":2116},[2117,2118,2119,2120],"Add guarded Crystal field Potion use for living injured Pokémon, with live menu verification, exact consumption checks and bounded menu cleanup.","Recover low-HP early Crystal starters via bounded wild-battle RUN inputs and an ordinary-input route from eastern Route 29 to Elm's healing machine.","Retain controller ownership through healing and lab exit; suppress policy navigation rewards and preserve the learned action\u002Fobservation contracts.","Include the 16.8.29 battle diagnostics in the Crystal release.",{"version":2122,"date":2113,"sections":2123},"16.8.29",[2124],{"name":1198,"changes":2125},[2126,2127,2128,2129,2130],"Rejoin Red's navigation pointer after the verified north-Forest recovery trip and successful Pewter healing, without granting skipped waypoint rewards.","Guide the Forest southwest clearing back to the eastern approach using collision-verified arrows and a shared, non-repeatable reward potential.","Decode actionable Crystal battle menus from live input routines and menu geometry, and report the move selected by a primitive A-button request.","Retain bounded final-battle traces with explicit entry, exit, truncation and completeness evidence; preserve Crystal blocked-item counters in death reviews.","Preserve Crystal 16.8.28 detour guidance and stale-drain telemetry fixes while retaining both games' learned weights, feature shapes and checkpoint lineages.",{"version":2132,"date":2113,"sections":2133},"16.8.28",[2134],{"name":121,"changes":2135},[2136,2137,2138,2139],"Give Crystal's first two waypoints verified detour-exit approach\u002Ftransition rewards while suppressing exploration rewards on incidental maps. Keep the policy in control and reward each approach high-water\u002Fexit identity once.","Scope both historical route projection and detour exploration suppression to the configured waypoint indices, preserving later Egg delivery and Johto quests.","Clear retained rollout policy tensors during inference-only episode drain; omit unavailable distributions and value estimates instead of recording stale data.","Preserve both learned lineages, checkpoint\u002Foptimizer tensors, observation\u002Faction shapes, the Crystal screen-feature bound, Nuzlocke rules and worker allocation.",{"version":2141,"date":2113,"sections":2142},"16.8.27",[2143],{"name":31,"changes":2144},[2145,2146,2147,2148],"Integrate the Red preparation encounter guard and Crystal recurrent feature bound, item-recovery diagnostics, Nuzlocke rules and Johto milestones into main.","Preserve the current stitched Crystal atlas and interactive analysis Sankey, including observed-location ribbons and recorded badge totals.","Retain each game's learned weights, optimizer state, worker allocation and checkpoint lineage. Live activation remains a managed version handoff.","Accept older preparation telemetry records that lack the new unsafe-encounter counter, so fallback diagnostics cannot interrupt an episode.",{"version":2150,"date":2113,"sections":2151},"16.8.26",[2152],{"name":1198,"changes":2153},[2154,2155],"Opt Crystal into an explicit screen-feature L2 bound before the recurrent layer. Preserve learned tensors, optimizer state, observation shapes and checkpoint lineage; persist the transform in saved policies and reject incompatible transform changes on resume.","Preserve blocked battle-item attempts in local episode metrics, CSV summaries and terminal capsules, with bounded context and explicitly censored recovery evidence. Report Crystal progress phases using Crystal milestones.",{"version":2157,"date":2113,"sections":2158},"16.8.25",[2159],{"name":121,"changes":2160},[2161,2162],"Release Red team preparation after eight distinct unsafe training encounters without a verified training KO, roster improvement, or level\u002Fcapability gain. Flee retries count once; healing and reordering cannot renew the budget.","Finish the current battle and allow recovery before the field handoff. Preserve legal recruitment, genuine blackout outcomes, learned weights and the six-worker lane.",{"version":2164,"date":2165,"sections":2166},"16.8.24","2026-09-12",[2167],{"name":1198,"changes":2168},[2169,2170],"Let Crystal episodes continue after the verified ROM guard cancels a forbidden battle-item attempt before consumption or effect. Report the episode count of blocked attempts in Crystal rule telemetry. Actual consumed-item violations, invalid catches and unavailable guards retain their existing enforcement.","Verify repeated Potion cancellation, menu exit and subsequent move execution with an ordinary-input real-ROM replay across environment resets.",{"version":2172,"date":2165,"sections":2173},"16.8.23",[2174,2178,2182],{"name":1804,"changes":2175},[2176,2177],"Crystal's 84 ordered Johto milestones, from the natural starter states to recorded Hall of Fame victory, including the Mystery Egg return, Floria detour, HM acquisition, gyms, Rocket quests and Elite Four.","Pinned Crystal story-flag decoding and exact Johto badge requirements. Scripted quest warps preserve earned progress; missing flags cannot be replaced by visiting a tile.",{"name":31,"changes":2179},[2180,2181],"Crystal episodes may continue to 500,000 steps while waypoint\u002Fdistance\u002Fbadge progress renews a 50,000-step timeout. The existing 50,000-step observation horizon, policy input shape, learned lineage and three-worker assignment remain compatible.","Both Elm lab exits advance the Egg-return route; stalled quest requirements remain visible in reward telemetry.",{"name":2183,"changes":2184},"Activation",[2185],"Prepared for the existing Crystal scheduler's next natural batch boundary, resuming a sealed checkpoint. This source release does not itself establish live activation.",{"version":2187,"date":2165,"sections":2188},"16.8.22",[2189],{"name":31,"changes":2190},[2191,2192,2193],"Run new ordinary Red batches with six workers, balanced two per starter, alongside the independently managed three-worker Crystal lane. Preserve the recurrent policy, optimizer settings, two-million-step budget and checkpoint lineage; retain validation of sealed twelve-worker batches and checkpoints.","Scope Red process ownership so the known Crystal starter trainer does not block Red reconciliation or startup. Unknown trainer commands still block.","Releases 16.8.20 and 16.8.21 belong to the isolated Crystal source branch; this Red release leaves its active process and installation untouched.",{"version":2195,"date":2165,"sections":2196},"16.8.21",[2197],{"name":31,"changes":2198},[2199,2200,2201],"Add separate OpenClaw Crystal training and cloud-sync checks matching Red's one-minute and five-minute cadence. Continue verified checkpoints between batches.","Keep Crystal at three workers, one per starter, with zero evaluation workers while Red has priority. Increasing Crystal allocation requires a later explicit decision after Red is beaten.","Adopt live processes through their existing locks, prevent overlapping launches, preserve the current lineage target during interruptions, and stop unsafe retries.",{"version":2203,"date":2165,"sections":2204},"16.8.20",[2205],{"name":31,"changes":2206},[2207,2208,2209,2210],"Upload Crystal training under the separate `pokemon-crystal` project with explicit game metadata, Johto starter status and collision-safe historical identities.","Enable Crystal Run Atlas and Run Analysis with game-scoped requests and caches, Johto map geometry, Generation II species, and observed-location analysis.","Enforce project ownership during Cloudflare ingestion so Crystal identifiers cannot overwrite Red runs, executions, episodes, status or artifacts.","Preserve Crystal's three starter workers, Nuzlocke rules, recurrent weights and `pokemon-crystal-starters-v1` lineage through managed release activation.",{"version":2212,"date":2165,"sections":2213},"16.8.19",[2214,2217,2223],{"name":2215,"changes":2216},"Red branch",[],{"name":31,"changes":2218},[2219,2220,2221,2222],"Treat exhausted team preparation as a soft warning, releasing the learned policy to explore for the rest of the episode instead of terminating it. Preserve normal progress timeouts, absolute episode limits, Nuzlocke rules, level caps and independent recovery\u002Fbattle safety.","Stop preparation-only retreats, readiness gates, recruitment and training after release. Finish already owned menu transactions within their existing bounds, then release failed result\u002Fpurchase cleanup instead of waiting forever.","Record the warning reason, release step and waypoint, subsequent steps and waypoint advance in compact diagnostics and episode exports.","Preserve compatible recurrent weights, optimizer state and checkpoint lineage; live activation remains a separate managed version handoff.",{"name":2224,"changes":2225},"Crystal branch",[2226,2227,2228,2229],"Enable Crystal permanent death, first-blackout termination, first eligible encounter per named area with evolution-family duplicates, forced Set mode, battle item restrictions, and Johto gym\u002FLeague level caps.","Use Crystal symbols and generation-two capture item IDs for enforcement; retain the three starter workers, observation\u002Faction contract, and compatible `pokemon-crystal-starters-v1` learned checkpoint lineage.","Preserve Crystal starter IDs and names in episode and cohort reports.","Isolate the code release from live activation through a checkpointed managed version handoff. Red and Silver configurations remain unchanged.",{"version":2231,"date":2165,"sections":2232},"16.8.18",[2233,2235,2240,2242],{"name":2215,"changes":2234},[],{"name":1198,"changes":2236},[2237,2238,2239],"Add Route 22's eastern grass as a bounded early recruitment fallback, with ordinary-input return to Viridian and Route 2, keeping policy rival-detour gates, first-encounter legality, catch safety and preparation step limits.","Recruit another legal teammate when too few living members can deal damage and the protected level ceiling blocks training a replacement fighter. Release recruitment after that shortage is resolved without resetting the preparation clock or changing the readiness targets.","Preserve compatible recurrent weights, optimizer state and checkpoint lineage. Source changes require managed version handoff before live activation.",{"name":2224,"changes":2241},[],{"name":1804,"changes":2243},[2244,2245,2246,2247],"Add an isolated English Crystal v1.0 runtime with pinned ROM\u002Fsymbol identity, Gen II memory decoding, CGB mode and a locked real-time clock.","Train one fresh shared recurrent Crystal policy across fixed Chikorita, Cyndaquil and Totodile worker cohorts. Validate natural reset-state provenance, actual starter identity and balanced assignments before policy control.","Provide startup checks and a three-worker bootstrap configuration with separate observations, checkpoints and logs. Full Crystal Nuzlocke menus, route\u002Fstory control and HM execution remain uncalibrated and unavailable.","Preserve the Red and Silver policy contracts. Source release registration does not install or activate a release in the managed Red trainer.",{"version":2249,"date":2165,"sections":2250},"16.8.17",[2251],{"name":1198,"changes":2252},[2253,2254,2255,2256],"Restore local Forest navigation guidance for supported stale objectives, preserving semantic prerequisites, encounter\u002Fdeath ledgers, and reward authority.","Route northbound Forest guidance through the clear aisle beside the uncollected Antidote, preventing a later item-blocked navigation stall.","Clear stale collision history after preparation releases navigation control, so successful entrance visits do not trigger a false blocked-route failure. Keep the existing bound for consecutive blocked movement.","Preserve compatible recurrent weights, optimizer state, and checkpoint lineage through the managed training version handoff.",{"version":2258,"date":2165,"sections":2259},"16.8.16",[2260],{"name":1198,"changes":2261},[2262,2263,2264],"Give emergency field medicine for a poisoned living party member priority over recovery travel, keeping bounded item-menu cleanup and ordinary navigation-first recovery intact.","Prioritize participation for a needed status-only teammate before spending helper XP on solo training. Diagnose capped paths to a missing damage-capable fighter without relaxing the protected level ceiling or team readiness requirements.","Restore Pewter Gym recovery coverage for the full-run lane and connect the west interior around its rock row to the existing exit and Center route.",{"version":2266,"date":2267,"sections":2268},"16.8.15","2026-09-11",[2269],{"name":1198,"changes":2270},[2271,2272,2273],"Clear the exact ready-team Pewter Gym landing at (4,13) with replay-verified UP input, preserving real transition guards and returning control to the policy after reaching the aisle.","Show observed mainline progress in full-run charts and run details when controller travel has advanced beyond the reward-eligible frontier. Keep training rewards and checkpoint selection unchanged.","Respect explicit terminal reasons before using lead HP as a legacy ending fallback, so a fainted lead does not turn a recorded stall into a whiteout.",{"version":2275,"date":2267,"sections":2276},"16.8.14",[2277],{"name":1198,"changes":2278},[2279,2280,2281],"Rejoin the Forest entrance route from the observed east landing tile, allowing ready teams to clear the corridor and return control to the policy instead of timing out. Keep transition and interaction input guards intact.","Add retained-state and controller-boundary regressions for the observed Forest stalls, preserving party state, Nuzlocke ledgers, and actor ownership.","Keep an active preparation escape ahead of proactive low-HP switching, while retaining mandatory replacement of a fainted carrier. Require known adequate carrier speed before selecting the bounded switch-and-escape plan.",{"version":2283,"date":2284,"sections":2285},"16.8.13","2026-09-10",[2286],{"name":1198,"changes":2287},[2288,2289],"Defer missing starter north-exit evidence for undersampled interrupted batches while preserving observed regression checks and completed-batch gates.","Add explicit managed recovery from a verified historical interrupted checkpoint, bound to its original launch and the next code handoff. Preserve optimizer state, the accepted-best checkpoint, and saved current-run artifacts.",{"version":2291,"date":2284,"sections":2292},"16.8.12",[2293],{"name":1198,"changes":2294},[2295,2296,2297],"Connect audited Center service-aisle tiles (4,4) and (4,6) to the shared exit corridor, keeping recruitment and preparation moving after healing.","Connect the Forest gate's southern Route 2 landing to the southbound recruitment and restocking corridor.","Add route coverage checks across recruitment goals and ordinary-input emulator regressions for the repaired Center and Forest connections.",{"version":2299,"date":2284,"sections":2300},"16.8.11",[2301],{"name":1198,"changes":2302},[2303],"Keep the Viridian and Pewter Pokémon Center PC exit corridor connected at tile (4,5), allowing recruitment and preparation to reach the exit after healing or depositing a fainted teammate.",{"version":2305,"date":2284,"sections":2306},"16.8.10",[2307],{"name":1198,"changes":2308},[2309,2310,2311],"Limit historical periodic checkpoint uploads to the two latest distinct numeric steps per execution across release directories. Preserve the current execution's saves, named models, and nonperiodic provenance artifacts.","Check the local retirement registry before artifact uploads and metadata ingestion, preventing removed immutable objects from being uploaded again. Invalid registry identities stop artifact uploads until corrected.","Drain retired upload requests with explicit skipped receipts and protect local checkpoints from cleanup, including those with older upload receipts. Preserve compatible model weights, optimizer state, and training lineage.",{"version":2313,"date":2284,"sections":2314},"16.8.9",[2315],{"name":1198,"changes":2316},[2317,2318,2319],"Preserve recorded catch locations after evolution, keep individual catch origins distinct in historical teams, and record captures after PC deposits even when the Pokédex already knows the species.","Apply the Nuzlocke duplicate clause to the entire evolutionary family for the run, including related forms of boxed, fainted, and initially owned Pokémon.","Record scripted starters as owned without awarding policy capture credit.",{"version":2321,"date":2284,"sections":2322},"16.8.8",[2323],{"name":31,"changes":2324},[2325],"Let the managed training cycle adopt a newer installed release at an optimizer boundary, finish active games within a bounded window, and retain sealed checkpoints and explicit version-upgrade outcomes.",{"version":2327,"date":2284,"sections":2328},"16.8.7",[2329],{"name":1198,"changes":2330},[2331],"Guide prepared teams through the southern Forest gate using the surveyed doorway route, with a bounded handoff that returns control to the policy.",{"version":2333,"date":2334,"sections":2335},"16.8.6","2026-09-09",[2336],{"name":1198,"changes":2337},[2338],"Continue team preparation with the living roster after a teammate is retired, preserving mandatory healing and permanent-death restrictions.",{"version":2340,"date":2334,"sections":2341},"16.8.5",[2342],{"name":31,"changes":2343},[2344],"Prepare living teams before the Forest and Pewter Gym using explicit readiness targets, bounded training, and verified final healing.",{"version":2346,"date":2334,"sections":2347},"16.8.4",[2348],{"name":1198,"changes":2349},[2350,2351],"Distinguish the opening Start menu from unfinished nurse dialogue, preventing a false healing timeout after a successful heal while preserving real dialogue recovery and permanent-death guards.","Report terminal trace coverage from actual exported action records rather than storage-reference flags, so death reviews identify missing evidence.",{"version":2353,"date":2354,"sections":2355},"16.8.3","2026-09-08",[2356],{"name":1198,"changes":2357},[2358,2359,2360,2361],"Guide Parcel-carrying runs back to Oak when physical travel has passed stale semantic anchors. Audited Pallet and Lab routes preserve progress until the real handoff, without changing bedroom resets or policy action ownership.","Stop queued nurse inputs after an observed poison faint and redirect to the PC, preserving permanent deaths and living party members.","Replan learned Forest recovery when a predicted map transition has not begun, retaining settling for real warps and adding the verified southern exit tile.","Stream the versioned Best Runs migration within D1's memory budget, preserving historical winners and allowing stalled completed-episode uploads to retry.",{"version":2363,"date":2364,"sections":2365},"16.8.2","2026-09-07",[2366],{"name":31,"changes":2367},[2368,2369,2370],"Prevent successive short movement inputs from being recorded as one false two-tile connection. Preserve observed hazards, existing ambiguous inputs, exact-context planning, and shared learned-map compatibility.","Enable existing low-HP and PP recovery through Route 2, both Forest gates, the Forest, and the northern Pewter nurse corridor, including a lone survivor after dead-party deposits.","Recognize automatic selection of a living lead during a battle introduction without pressing through the command menu into an unintended attack. Keep mandatory replacement of permanently retired members and controller ownership.",{"version":2372,"date":2364,"sections":2373},"16.8.1",[2374],{"name":121,"changes":2375},[2376,2377,2378],"Resolve mandatory recovery switching during battle introductions when a permanently retired party member has been healed in game memory. Preserve the death ledger and select a living replacement through ordinary controls before resuming escape, including when proactive battle safety is disabled.","Repair early Viridian guidance and doorway movement using retained failure replays. Keep actual story transitions and waypoint credit authoritative.","Preserve 16.8.0 action filtering, frontier reward timing, learned-map sharing, learning settings, recurrent weights and optimizer continuation. Validate the repair in isolated emulators before the next ordinary batch launch.",{"version":2380,"date":2381,"sections":2382},"16.8.0","2026-09-06",[2383],{"name":121,"changes":2384},[2385,2386,2387,2388],"Filter unavailable actions before recurrent PPO sampling and use the same stored availability for PPO updates. Preserve the ten output heads, existing checkpoint weights and optimizer, useful healing\u002Fstory\u002Fdoorway interactions, and controller ownership. Support ordinary checkpoint and evaluation loaders.","Credit verified policy-owned frontier progress on its actual step, once per advance. Retain terminal failure penalties without a delayed progress payout; exclude invalid and controller-owned advances and retain legacy timing as an independent configuration option.","Share compatible observed recovery connections in a bounded planning view of the current exact game context. Preserve local maps, shared negative evidence, conservative risk, pinned seeds and authored fallback; expose sharing and fallback diagnostics. Do not combine every context into an oversized map.","Enable all three options for the next ordinary Red progress batch. Verify action availability, recurrent probability accounting, reward timing, shared recovery replays and checkpoint continuity in isolated diagnostics.",{"version":2390,"date":2381,"sections":2391},"16.7.2",[2392],{"name":121,"changes":2393},[2394,2395,2396,2397],"Carry pending permanent-death deposits toward a PC throughout recovery. Leave heal-only homes, use a PC approach through Pallet, and retain the existing shared corridors and bounded failure guards, including blocked doorways.","Allow a complete learned PC path to leave a heal-only home. Record learned inputs, learned transition settling, and authored inputs separately in each episode so map growth is not mistaken for learned-route usage.","Match battle PP and HP comparisons by stable party identity across switches and battle exits. Report unavailable comparisons without inventing damage or move consumption.","Retain switch outcomes, retry flags, action ownership, terminal outcomes, and full-party identity\u002FPP in bounded battle traces. Report actual trace eviction even for recovery-owned battle steps.",{"version":2399,"date":2381,"sections":2400},"16.7.1",[2401],{"name":121,"changes":2402},[2403,2404,2405,2406],"Recover around the uncollected Forest Antidote using a verified alternate direction field after movement is blocked. Preserve ordinary route timing, menu cancellation, and the existing stuck\u002Fprogress limits.","Recheck the live bag after catch-menu synchronization so a pending last Poké Ball throw cannot shift a Potion into a cached Ball slot.","Attribute forced recovery escapes to the controller even when the sampled policy action was masked, keeping those actions out of PPO actor loss.","Replay both blocked Forest failures and the recorded illegal-item failure; preserve the earlier recovery replays and learned-routing safeguards.",{"version":2408,"date":2381,"sections":2409},"16.7.0",[2410],{"name":121,"changes":2411},[2412,2413,2414,2415],"Learn directed navigation from executed gameplay inputs and remember verified healing\u002FPC destinations. Prefer complete learned recovery paths, retaining existing routes when observed connectivity is incomplete.","Keep map collection passive. Confirm doorway arrivals and ledge landings, separate story\u002Fcapability contexts, exclude trainer hazards and ambiguous movement, and retain conservative encounter and HP-loss costs.","Persist bounded maps per worker with game\u002FROM identity checks and an audited, hash-pinned Red seed. Keep Center interactions, permanent-death enforcement, semantic progress suspension, and controller actor-loss masking intact.","Verify recovery from a withheld exploration branch after removing handwritten overworld routes, plus identical emulator frames and RAM with collection on\u002Foff.",{"version":2417,"date":2418,"sections":2419},"16.6.2","2026-09-05",[2420],{"name":121,"changes":2421},[2422,2423,2424],"Repair RUN synchronization when the battle cursor becomes actionable before its menu finishes drawing. Cancel nested move menus without confirming an attack, spending move PP, or using a forbidden item.","Replace Forest recovery waypoint joins with direction fields built from verified legal movement, including both gate exits. Reject unknown paths instead of snapping across trees, and retain bounded stuck and progress checks with a movement retry after menu cancellation.","Replay the recorded flee failure and all four recovery failures in isolated emulators, alongside older route joins and permanent-death safeguards.",{"version":2426,"date":2418,"sections":2427},"16.6.1",[2428],{"name":121,"changes":2429},[2430,2431,2432,2433,2434],"Repair Viridian Forest recovery joins using verified routes around blocked tiles, retaining the existing no-progress guard and permanent-death rules.","Add direction fields for Forest waypoints 62–66 and correct the north doorway's exact transition source to `(1,1)`. Verify legal movement and zero net navigation reward on repeated loops.","Enable legal field Antidote treatment before movement, verifying live menus, exact item consumption, and the same living target. Preserve retired party members, record controller ownership, and bound failed-menu retries without preemptively ending poisoned episodes.","Prevent catch-menu synchronization and raw confirmations from consuming forbidden battle items while preserving legal Poké Ball use.","Guide Viridian southbound Parcel and post-delivery returns through a collision-audited direction field, including wall corners and route rejoins.",{"version":2436,"date":2418,"sections":2437},"16.6.0",[2438],{"name":121,"changes":2439},[2440,2441,2442,2443,2444],"Add explicit Pokémon Red and Pokémon Silver game profiles while preserving Red as the default for every historical config and checkpoint.","Decode standard English Silver state through a pinned pret\u002Fpokegold symbol table, including banked maps, Gen II party records, inventory pockets, badges, Pokédex, battle state, happiness, and split Special stats.","Put PyBoy construction behind an emulator backend boundary, lock Silver's RTC, and route training, evaluation, state capture, and inspection through profile-aware factories.","Add a fail-closed Silver preflight that verifies the ROM and symbol hashes, required labels, initial state, CGB\u002FRTC settings, safe controllers, output isolation, and fresh eight-action policy lineage before workers start.","Add a ROM-independent Silver bootstrap config with conservative exploration rewards, profile-scaled Pokédex accounting, and capped decaying wild-battle credit. Red navigation and reward settings are unchanged; Silver story and Nuzlocke automation remain gated on replay calibration from the real ROM.",{"version":2446,"date":2418,"sections":2447},"16.5.1",[2448],{"name":121,"changes":2449},[2450,2451,2452,2453],"Replace waypoint 54's narrow Viridian corridor with a legal-input PyBoy direction field covering all 600 audited city positions, including the dynamic Center-door edge found in a 16.5.0 failure capsule. Unknown collision geometry now produces no target instead of a straight line through a wall.","Shape each corridor move with a signed path-distance delta: forward movement earns credit, retreat loses the same credit, and loops net zero. Keep the episode high-water mark as the sole distance evidence for lease renewal.","Restore waypoint 54 to the global 16,000-step stall timeout and allow only route progress, semantic completion, or verified healing to renew that phase.","Require one proven Viridian-to-Route-2 transition from every starter cohort before a sealed 16.5.1 batch checkpoint can advance the training lane.",{"version":2455,"date":2456,"sections":2457},"16.5.0","2026-09-04",[2458],{"name":121,"changes":2459},[2460,2461,2462,2463,2464,2465,2466,2467],"Guide the policy through forward-replayed Route 2 and early Viridian Forest corridors, including the ledge gap, southwest pocket, sign, and trainer detours. Correct the Forest gate's exact transition source to `(3,44)`.","Rejoin local corridors after healing while preserving semantic progress. Count only new best forward corridor distance, preventing retreat-and-rejoin loops from replaying navigation rewards or extending the progress lease.","Give new ordinary batches two million optimizer transitions with the same 250,000-transition checkpoints and twelve-worker pool. Seal each launch's budget and retain verification of historical one-million-transition batches.","Extend the local 24,000-step progress floor through waypoints 56–61 while retaining the global 16,000-step floor. Preserve the learning rate, entropy, frozen encoder, action space, and observation shape.","Automatically retry an ordinary batch after an unclean exit before its first checkpoint, using the unchanged verified launch parent. Preserve the abandoned execution and checkpoint records, and continue to alert on failed verification.","Project waypoint features and dense navigation potential through shared, collision-audited Pallet-to-Lab and Viridian-to-Route-2 route anchors without changing semantic waypoint indices or executing controller movement.","Exclude controller-owned recurrent transitions from actor, entropy, clipping, and KL losses while retaining them for critic learning.","Keep the shared visual encoder frozen behind balanced, checkpoint-bound policy-owned frontier evidence and publish exact per-starter waypoint sample counts for that gate.",{"version":2469,"date":2456,"sections":2470},"16.4.0",[2471],{"name":2472,"changes":2473},"My Pokémon run history",[2474,2475,2476],"Add a My Pokémon section to run details showing teammates throughout an attempt, including fainted Pokémon later deposited in the PC.","Preserve individual Pokémon history across party switches and evolution, with party\u002FPC status and the recorded location of each Pokémon's faint.","Use available starter, party, and catch records for older attempts, clearly identifying missing status and faint locations.",{"version":2478,"date":2479,"sections":2480},"16.3.5","2026-09-03",[2481],{"name":2482,"changes":2483},"Pre-Parcel route protection",[2484,2485,2486],"Route every optional wild battle before Parcel delivery through the verified flee controller, with its last-resort attack fallback disabled so encounters cannot become an eight-win grinding loop.","Guide the policy down a collision-replayed Route 1 Parcel-return corridor and through the Forest south gate, while synchronizing stale semantic pointers only after the emulator proves each exact route transition.","Reject a completed batch before baseline collection when Parcel acquisitions fail to convert into deliveries, including a balanced per-starter check, so a locally regressed optimizer checkpoint cannot become the next resume point.",{"version":2488,"date":2489,"sections":2490},"16.3.4","2026-09-02",[2491],{"name":2492,"changes":2493},"Affordability-aware Poké Ball reserves",[2494,2495,2496],"Raise the opening reserve target from five to ten Poké Balls and, on later visits to an audited Mart, scale it to 15, 20, and 25 after badge milestones.","Restock only to the quantity current cash can actually cover, and stop a clerk loop after repeated no-progress transactions.","Enable the same verified purchase flow in the historically observed Pewter Mart, allowing another restock before Route 3 and Mt. Moon encounters.",{"version":2498,"date":2489,"sections":2499},"16.3.3",[2500],{"name":2501,"changes":2502},"Restock-safe Mart continuity",[2503,2504],"Restore unrestricted Viridian Mart entry so a depleted party can legally restock Poké Balls after its Route 1 encounter.","Convert only clerk interactions at or above the configured Poké Ball target into policy-owned idle actions, discouraging the repeated shopping loop without blocking legitimate purchases.",{"version":2506,"date":2489,"sections":2507},"16.3.2",[2508],{"name":2509,"changes":2510},"Post-encounter Route 2 continuity",[2511,2512],"Align stale post-Parcel waypoint guidance to the Viridian return phase only after the emulator proves the Route 1 encounter was spent and the exact northbound Route 1 transition was crossed.","Block pre-Brock Mart re-entry after that return, preventing the policy from replaying its learned shopping loop instead of taking Viridian's north exit.",{"version":2514,"date":2489,"sections":2515},"16.3.1",[2516],{"name":2517,"changes":2518},"Pre-Forest level-cap guard",[2519,2520,2521],"Escape optional wild battles once a party member reaches one level below the active badge cap, while always preserving a legal policy-requested catch.","Stop wild victories from renewing the continuous-run progress lease and clamp a level-cap failure's complete episode return to a negative value.","Close only Viridian's southbound Route 1 transition after the catch loop has returned to town, leaving the required initial trip and later badge-gated travel available.",{"version":2523,"date":2489,"sections":2524},"16.3.0",[2525],{"name":2526,"changes":2527},"Balanced 12-worker full-run collection",[2528,2529,2530],"Add a versioned ordinary-lane worker-pool contract with twelve training environments allocated 4\u002F4\u002F4 across Bulbasaur, Charmander, and Squirtle.","Preserve the 512-step recurrent horizon and 768-sample PPO minibatch while increasing each rollout from 4,608 to 6,144 transitions.","Keep in-flight nine-worker batches reconcilable, require new batches to bind the 12-worker config and contract digests, and provide an isolated one-rollout checkpoint-resume canary before production activation.",{"version":2532,"date":2533,"sections":2534},"16.2.0","2026-09-01",[2535],{"name":2536,"changes":2537},"Parcel-delivery route continuity",[2538,2539],"Treat verified removal of Oak's Parcel in his Lab as authoritative route evidence, monotonically synchronizing stale geometric guidance to the post-delivery phase without issuing inputs or granting skipped rewards.","Require each starter cohort to demonstrate at least three verified deliveries and 75% delivery-to-post-delivery alignment before curriculum promotion, isolated to fresh 16.2 evidence.",{"version":2541,"date":2533,"sections":2542},"16.1.0",[2543,2549,2554],{"name":2544,"changes":2545},"Battle-menu synchronization and checkpoint safety",[2546,2547,2548],"Require matching battle RAM and framebuffer evidence before issuing a wild- battle command, and restore a neutral input edge while following cursor transitions so a stale pre-battle cursor cannot misroute RUN.","Retain invalid flee-synchronization outcomes in the incremental health cursor, fail the active batch's gate immediately when one occurs, and cancel any semantic checkpoint promotion when that gate rolls back.","Bind health reconciliation to the completed batch's proven release instead of the scheduler's newly installed release, preserving correct validation across an in-flight minor-version upgrade.",{"name":2550,"changes":2551},"Operational ownership and diagnostics",[2552,2553],"Add deterministic replay and trace support for bounded failure capsules; the captured flee-synchronization capsule now completes all 94 requested steps without a synchronization abort.","Project full-run status from the ordinary cycle owner, report the historical V20 guardian as a non-owner, and correct the local status-job manifest to use that operational source of truth.",{"name":2555,"changes":2556},"Bounded training observability",[2557,2558,2559,2560],"Add a compact incremental episode index with general waypoint conversion, per-starter story funnels, first-hit timing, 64\u002F128\u002F512-episode statistical previews, and ranked failure fingerprints so routine diagnosis no longer rescans multi-gigabyte episode history.","Add bounded failure flight recorders paired with rolling diagnostic-only PyBoy checkpoints, sparse full policy action distributions, and an isolated capsule replay helper.","Attribute PPO rewards and computed advantages by story phase, starter, and policy\u002Fcontroller\u002Frejection ownership without changing the rollout buffer or optimizer.","Report Cloudflare spool pressure, queue age, uploader freshness, and stable upload-failure categories; divert sampled traces into a bounded, restart-recoverable local emergency ring when the spool is saturated instead of silently dropping them.",{"version":2562,"date":2563,"sections":2564},"16.0.0","2026-08-31",[2565],{"name":2566,"changes":2567},"Opening rival training continuity",[2568,2569,2570],"Treat exactly one fully proven Oak's Lab opening-rival whiteout as a training-only, nonterminal event: omit its faint and blackout penalties, keep the starter out of the permanent-death ledger, advance the ordinary defeat dialogue and heal\u002Fwarp, then resume the same episode.","Require the exact map, waypoint, first-battle state, single level-5 base starter, matching level-5 rival starter, zero badges and wins, and pre-Parcel story state; every near miss and every later death retains normal Nuzlocke penalties and termination.","Disable the exception in held-out evaluation and record dedicated ignored- loss telemetry so evaluation outcomes and training continuation remain independently auditable.",{"version":2572,"date":2573,"sections":2574},"15.1.9","2026-08-30",[2575],{"name":2576,"changes":2577},"Viridian northbound and flee synchronization",[2578,2579],"Align the post-capture Viridian north-exit waypoint with the collision-safe x=18 approach and accept both legitimate north warp tiles, preventing a Route 2 transition from being entered without advancing the ordered route.","Normalize the wild-battle command cursor from exact live RAM after every bounded movement, recovering a dropped directional input while submitting RUN only after its exact cursor position is proven.",{"version":2581,"date":2573,"sections":2582},"15.1.8",[2583],{"name":2584,"changes":2585},"Forest and wild-battle controller recovery",[2586,2587],"Route Viridian Forest's southwest pocket through a collision-replayed south-gate suffix instead of snapping toward a geometrically close waypoint across an impassable wall, while retaining narrow branch selection around wall-separated starts.","Re-confirm a dropped wild-battle RUN command once with the robust button hold only when live menu RAM proves the cursor is still exactly on RUN, then retain the existing bounded fail-closed synchronization abort if the retry is not accepted.",{"version":2589,"date":2590,"sections":2591},"15.1.7","2026-08-29",[2592],{"name":2593,"changes":2594},"Viridian Route 22 detour guard",[2595,2596],"Mask the four westbound Viridian-to-Route-22 warp tiles until eight badges, matching the existing Route 22 frontier phase while leaving the northbound Route 2 objective and late-game return fully policy-owned.","Attribute blocked westbound requests to the policy, apply the existing no-op\u002Frepeat shaping, and retain row-specific gate counts in episode telemetry so boundary stalls remain measurable.",{"version":2598,"date":2590,"sections":2599},"15.1.6",[2600,2607],{"name":2601,"changes":2602},"Stall recovery and tactical intent",[2603,2604,2605,2606],"Record whether each unsafe opening-rival controller attack aligns with the policy's sampled attack head while retaining legal low-HP finisher behavior.","Use four Growls for Bulbasaur against the opening Charmander after the exact 999,972-step continuation's canonical-bedroom panel won 30\u002F36 versus 28\u002F36 for two Growls, improving Bulbasaur from 8\u002F12 to 10\u002F12 with no other-starter, item-legality, or move-execution regression.","Keep mandatory post-faint party replacement active when proactive low-HP switching is disabled, preventing recovery from looping indefinitely on a forced-switch menu instead of selecting a living teammate.","Report explicit reasons for masked Potion, trainer-catch, and trainer-item confirmation requests, and distinguish policy-aligned unsafe rival attacks from unrequested forced attacks without changing the ten-action checkpoint.",{"name":2608,"changes":2609},"Route 2 continuation visibility",[2610,2611],"Treat the full-run ordered Viridian-to-Route-2 transition waypoint as exact Route 2 progress in web run analysis even when the conservative rewardable map frontier remains at Viridian Mart, and report how many post-delivery stalls reached Route 2 before ending.","Extend the ordinary v3 lane's 24,000-step rolling lease floor from Parcel delivery through the first Route 2 anchor while preserving the 16,000-step global zero-badge timeout and the isolated v4\u002Fv5\u002Fv6 continuation contracts.",{"version":2613,"date":2614,"sections":2615},"15.1.5","2026-08-28",[2616],{"name":2617,"changes":2618},"Monotonic parcel-learning continuation",[2619,2620],"Authorize the exact canonical bedroom state for internal promotion evidence, so the per-starter Parcel conversion gate counts real clean-start episodes.","Continue completed batches from their final sealed periodic checkpoint and recover that checkpoint if a mutable `latest.zip` alias is replaced before the scheduler evaluates it, preventing accidental PPO-lineage rollback.",{"version":2622,"date":2623,"sections":2624},"15.1.4","2026-08-27",[2625],{"name":2626,"changes":2627},"Verified Oak Parcel handoff",[2628,2629,2630],"Let the existing policy-selected `USE_POTION` head trigger a bounded Oak interaction only from the exact delivery waypoint, prove facing through the occupied Oak tile, and require the real Parcel removal plus cleared dialogue before reporting success.","Retain interaction attempts, orientation proof, A presses, Parcel removal, dialogue cleanup, and failure reasons in internal episode telemetry without adding new fields or stages to the public dashboard.","Withhold promotion until each starter converts at least 50% of eligible current-release Parcel acquisitions into deliveries, while preserving the existing 24,000-step post-Parcel progress lease.",{"version":2632,"date":2623,"sections":2633},"15.1.3",[2634],{"name":2635,"changes":2636},"Bounded v6 Viridian handoff",[2637,2638,2639,2640,2641],"Extend the local 24,000-step progress lease to waypoint 54, covering the measured Viridian north-exit bottleneck as well as Route 2 and the immediate Forest handoff, without widening the global zero-badge lease.","Queue one exact-parent, 500,000-step v6 tranche behind the active v3 batch; the scheduler seals v3 before launch and grants no automatic extension or promotion authority.","Freeze the current 512-episode v3 health baseline and stop v6 behind starter-balanced Forest-entry, Parcel-delivery, rival-safety, legal-item, and move-execution evidence gates.","Bind the handoff to the fully resolved inherited configuration, package source, scheduler source, and exact parent so drift fails closed before launch or terminal evaluation.","Add a recoverable handoff-retirement path: preserve the stopped v6 checkpoint, telemetry, and log with hashes, clear only the active handoff slot, and resume ordinary v3 learning from its verified continuation.",{"version":2643,"date":2644,"sections":2645},"15.1.2","2026-08-26",[2646],{"name":2647,"changes":2648},"Recurrent v6 Route 2-to-Forest handoff",[2649,2650,2651,2652],"Keep the global zero-badge meaningful-progress lease at 16,000 steps while applying a 24,000-step floor only to the Route 2 approach, Forest South Gate, and immediate Forest-entry waypoints.","Synchronize stale geometric waypoints after exact Viridian-to-Route 2 and Route 2-to-Forest transitions only when Parcel delivery, Ball-era encounter activation, the spent Route 1 encounter, and the configured doorway all agree; skipped waypoint rewards remain uncredited.","Report Forest entries per million completed episode transitions overall and by starter in ordinary-cycle health summaries.","Declare the isolated, untrained `nuzlocke-full-run-recurrent-v6` continuation lane with the same exact hashed v3 parent checkpoint inherited through v5; this release records no v6 checkpoint, panel pass, promotion, or live launch.",{"version":2654,"date":2644,"sections":2655},"15.1.1",[2656],{"name":2657,"changes":2658},"Recurrent v5 active-HP reconciliation",[2659,2660,2661],"Reconcile a transient zero active-battle HP byte against the indexed party slot for battle eligibility, tactical waits, safety, finishing attacks, and damage advice while preserving both raw values in telemetry.","Require a zero party slot, permanent-death record, or visible replacement menu before treating the active member as fainted; forbidden battle items and the recurrent observation contract remain unchanged.","Declare the isolated, untrained `nuzlocke-full-run-recurrent-v5` continuation lane with the same exact hashed v3 parent checkpoint inherited through v4; this release records no v5 checkpoint, panel pass, promotion, or live launch.",{"version":2663,"date":2644,"sections":2664},"15.1.0",[2665],{"name":2666,"changes":2667},"Recurrent v4 handoff",[2668,2669,2670,2671,2672],"Add an isolated `nuzlocke-full-run-recurrent-v4` continuation lane rooted at the immutable v3 499,986-step checkpoint, with an exact path, size, and SHA-256 resume binding and unchanged recurrent observation\u002Faction contract.","Bound the first training tranche to 500,000 transitions with a `5e-6` learning rate, `0.001` entropy coefficient, one PPO epoch, preserved optimizer state, and the shared feature extractor frozen.","Promote only canonical bedroom starts, retain forbidden battle-item masking, and use the validated four-Growl Bulbasaur opening-rival setup.","Correct critical-reply attack safety and retain low-HP recovery-classified controller turns in terminal battle traces.","Pass the 36-attempt paired rival gate at 30 wins versus 27 for v3, paired Route 1 starter smoke at 3\u002F3 in both lanes, and canonical-bedroom smoke at 3\u002F3 with no blackouts, rejected battle actions, or move mismatches.",{"version":2674,"date":2675,"sections":2676},"15.0.12","2026-08-25",[2677],{"name":2678,"changes":2679},"Story-first Parcel delivery",[2680,2681,2682,2683],"Remove pre-Parcel wild-battle damage, victory, and level-up rewards, add a per-step wild-battle cost, and reserve progress-lease renewal for navigation, story events, and required trainer battles.","End an unproductive pre-Parcel detour on its eighth optional wild win and clamp the failed episode's complete return to `-500` so earlier navigation rewards cannot make the grinding trajectory profitable.","Retain bounded post-delivery wild-battle shaping while applying one shared reward cap across damage and victory signals.","Fail full-run health validation on any level-cap episode or excessive pre-Parcel wild wins, and support audited quarantine plus immutable-root restart of unhealthy optimizer continuations.",{"version":2685,"date":2675,"sections":2686},"15.0.11",[2687],{"name":2688,"changes":2689},"Opening recovery and verified healing",[2690,2691,2692],"Route an injured opening-rival survivor from Oak's Lab to Red's house and restore the living party through Mom's real dialogue interaction before the first Route 1 grass exposure.","Enable the audited Pallet–Route 1–Viridian recovery corridor while retaining standard Pokémon Center healing at Viridian; both facilities verify the resulting party health and refuse to revive permanent Nuzlocke deaths.","Add saved-state emulator replays for both the Pallet Home and Viridian Pokémon Center paths, including collision-safe Pallet doorway navigation.",{"version":2694,"date":2695,"sections":2696},"15.0.10","2026-08-24",[2697],{"name":2698,"changes":2699},"Production-grounded Pallet rival rollback",[2700,2701,2702,2703],"Restore the two-Growl Bulbasaur plan after completed production episodes measured 6\u002F9 rival wins on 15.0.8 versus 5\u002F15 on the four-Growl 15.0.9 release; the larger setup was a live regression despite passing the original frozen post-starter panel.","Add a bedroom-start rival panel so validation now includes starter creation, variable starter stats, scripted cohort selection, and the full pre-battle RNG path instead of relying only on one frozen post-starter state.","In the new panel, two Growls passed at 30\u002F36 wins versus 29\u002F36 for four, including 11\u002F12 versus 10\u002F12 Bulbasaur wins with unchanged 10\u002F12 Charmander and 9\u002F12 Squirtle results, 289\u002F289 matched controller moves, and zero rejected battle actions.","Reject an attempted damage\u002FHP-adaptive setup rule after it regressed to 28\u002F36 wins versus the four-Growl arm's 29\u002F36 in the same bedroom panel.",{"version":2705,"date":2695,"sections":2706},"15.0.9",[2707],{"name":2708,"changes":2709},"Bulbasaur rival variance reduction",[2710,2711],"Extend the frozen Bulbasaur setup sweep through six Growls. Four Growls won 8\u002F12 fixed-seed battles versus 7\u002F12 for the 15.0.8 two-Growl plan; six tied at 8\u002F12 but required extra setup turns, while five regressed to 5\u002F12.","Promote the four-Growl schedule for Bulbasaur against the opening Charmander while retaining live cursor verification, zero battle items, and the critical-aware forced attack path.",{"version":2713,"date":2695,"sections":2714},"15.0.8",[2715],{"name":2716,"changes":2717},"Pallet rival strategy and move-menu correctness",[2718,2719,2720,2721,2722],"Model Generation I critical-hit probability and damage separately from the normal damage path, including doubled critical level and unmodified combat stats, while retaining the unchanged recurrent observation contract.","Navigate the live one-based move-menu cursor instead of trying to normalize a wrapping menu with repeated Up inputs or verifying against the previous confirmed move.","Keep legal damaging-move authority through unsafe rival states and use a starter-scoped two-Growl Bulbasaur setup against Charmander; battle items remain forbidden.","Add a frozen Bulbasaur setup sweep: two Growls won 7\u002F12 battles versus 4\u002F12 with one and 6\u002F12 with three, with 429\u002F429 matched move executions.","Expand the frozen release panel to 12 seeds per starter. The 15.0.8 candidate passed at 30\u002F36 wins versus 26\u002F36 for the advisor baseline, including 7\u002F12 Bulbasaur, 11\u002F12 Charmander, and 12\u002F12 Squirtle wins, with 329\u002F329 matched controller moves and zero rejected battle actions.",{"version":2724,"date":2695,"sections":2725},"15.0.7",[2726],{"name":2727,"changes":2728},"Opening-rival tactical safety and battle observability",[2729,2730,2731,2732,2733,2734],"Add a read-only Generation I battle calculator and compact live advisor telemetry without changing the recurrent policy observation contract.","Mask forbidden trainer-battle item and catch inputs at execution source, and verify controller move submissions against Red's dedicated move-list cursor.","Add an opening-Lab-rival safety controller that uses only legal battle-menu inputs and records intended-versus-executed move evidence for every turn.","Add a frozen, starter-balanced, inference-only rival comparison panel; the 15.0.7 candidate passed at 6\u002F9 wins versus 5\u002F9 for the advisor-only baseline, with 59 verified controller turns and zero execution mismatches or rejected battle actions.","Preserve inherited evaluation configs as both raw and resolved frozen artifacts so replay and Cloudflare metadata uploads remain reproducible.","Rotate oversized legacy step logs atomically when telemetry columns change, preserving history without copying a 24 GB CSV during trainer startup.",{"version":2736,"date":2737,"sections":2738},"15.0.6","2026-08-21",[2739],{"name":2740,"changes":2741},"V20 execution-lock schema admission",[2742,2743],"Admit V19 and V20 to the existing inode-bound truthful execution-lock schema used by prior frozen-root recovery generations.","Keep the exact V17 lease preimage, runtime-config canonicalization, and all one-shot development-only limits unchanged.",{"version":2745,"date":2737,"sections":2746},"15.0.5",[2747],{"name":2748,"changes":2749},"V19 execution-lock binding correction",[2750,2751],"Bind the replacement launch to the exact unheld V17 execution-lease bytes left by the interrupted V17 execution, instead of the older V13 lease.","Preserve the V18 runtime-config canonicalization fix and all existing one-shot development-only limits.",{"version":2753,"date":2737,"sections":2754},"15.0.4",[2755],{"name":2756,"changes":2757},"V18 runtime-attestation correction",[2758,2759],"Supersede the interrupted, unbound V17 execution after its guardian compared the source seed and YAML key types against the deterministic JSON runtime.","Canonicalize the expected runtime config and derive the execution-rotated seed before binding, while retaining the immutable V13 root and scope.",{"version":2761,"date":2737,"sections":2762},"15.0.3",[2763],{"name":2764,"changes":2765},"V17 scheduler-binding correction",[2766,2767],"Supersede the unbound V16 activation after its stale scheduler-source digest blocked binding; no trainer launch or optimizer transition was consumed.","Bind the exact V17 cron manifest and reconciler digests before activation, then retain the same frozen checkpoint, optimizer state, and training scope.",{"version":2769,"date":2737,"sections":2770},"15.0.2",[2771],{"name":2772,"changes":2773},"Isolated V16 recovery launch",[2774,2775],"Replace the consumed zero-transition V15 launch with one bounded V16 launch from the same immutable V13 terminal checkpoint and unchanged optimizer and training settings.","Verify V16 authorization and root records without importing unpinned sibling modules inside the trainer's isolated Python runtime.",{"version":2777,"date":2737,"sections":2778},"15.0.1",[2779],{"name":2780,"changes":2781},"V15 recovery after the zero-transition V14 launch failure",[2782,2783,2784],"Admit the completed V13 terminal checkpoint through a distinct generic, create-only V15 frozen-root attestation after the V14 launch failed before execution binding or optimizer progress.","Add a fresh scheduler-owned V15 one-shot recovery generation with 737,280 optimizer transitions, nine 3\u002F3\u002F3 starter environments, preserved optimizer state, and no evaluation, qualification, retention, promotion, retry, fallback, or extension authority.","Fail closed on the consumed V14 attempt and preserve every V14 lifecycle record as immutable history.",{"version":2786,"date":2737,"sections":2787},"15.0.0",[2788],{"name":2789,"changes":2790},"Verified V14 training generation",[2791,2792,2793],"Preserve the exact completed V13 terminal checkpoint at lineage 144,762,678 as an immutable, content-addressed initialization root for one new bounded development-only generation. V13 remains consumed history and grants no retry, evaluation, panel, qualification, retention, or promotion authority.","Advance the sole scheduled guardian to recovery generation V14 with one optimizer-preserving 737,280-transition execution, nine training environments split 3\u002F3\u002F3 across starters, zero inline evaluation, and the unchanged first-controllable-bedroom-to-Hall-of-Fame episode contract.","Require create-only release-transition, authority, scheduler-binding, launch, process-activation, and execution-binding evidence. Reuse the exact existing guardian and status jobs, and fail closed on source, scheduler, process, checkpoint, budget, or identity drift without retry or fallback.",{"version":2795,"date":2796,"sections":2797},"14.0.26","2026-08-20",[2798],{"name":2799,"changes":2800},"Process-visible launch binding",[2801,2802,2803,2804,2805],"Preserve the consumed 14.0.25 v2 recovery claim as immutable, poisoned historical evidence. Its trainer was spawned but never activated or bound: the macOS process-visible interpreter replaced the claimed virtualenv `argv[0]`, so the fail-closed guardian handed the process off before any execution manifest, PPO update, or training transition was admitted.","Treat cloud execution identity `35637737-a383-4cea-ac4d-1e88221b0592`, written during that refused startup, as an orphan telemetry pointer, never as a candidate, execution binding, checkpoint, or evidence of training. V2 has zero admitted executions and zero transitions; its claim cannot be retried.","Authorize a distinct v3 development-only generation from the same exact sealed 2,999,997-step root. Bind the logical launcher and an independently measured process-visible interpreter identity before launch, then reuse that exact binding for activation, active-runtime validation, and emergency handoff. Launch the trainer in isolated interpreter mode with exact module paths. In a two-phase gate, the trainer first validates the claim and exact source\u002Fconfig\u002Fauthority\u002Fguardian\u002Froot handshake; the guardian then binds the exact Screen → login → bash → Python ancestry plus the `tee` sibling and atomically publishes activation, which the trainer must acknowledge before it may write cloud identity, create an execution manifest, or initialize environments.","Scope the handshake as fail-closed operational safety against accidental launch, configuration, and process drift in the managed lane. It is not authentication or containment against a hostile process running as the same operating-system user.","Reserve `1005200001`–`1005200100` for the separate v3 development panel. The v2 `1005100xxx` panel remains unused historical scope, and `100300xxx` remains untouched for a separately authorized future qualification.",{"version":2807,"date":2808,"sections":2809},"14.0.25","2026-08-19",[2810],{"name":2811,"changes":2812},"Semantic-retention recovery",[2813,2814,2815,2816],"Preserve the terminal 14.0.24 semantic-quality hold and its rejected branch as immutable evidence. Start a new development-only recovery identity from the sealed 2,999,997-step checkpoint at lineage 142,015,509; the rejected branch tip and its 4M\u002F5M descendants are never eligible resume fallbacks.","Preserve the selected checkpoint's optimizer while reducing value-loss pressure and freezing its shared feature extractor for one bounded 1,004,544-transition tranche. Qualification, retention certification, and promotion authority remain false.","Consume the recovery authority before process creation, then bind the exact process start, command, execution manifest, health identity, and sealed terminal checkpoint pair. Interrupted, drifting, unbound, overshooting, or repeated execution fails closed and cannot be retried.","Distinguish a verified terminal semantic hold from a stale live trainer, retain last-runtime provenance, and independently verify the existing public Cloudflare run-status readback without making telemetry availability a launch authority.",{"version":2818,"date":2808,"sections":2819},"14.0.24",[2820],{"name":2821,"changes":2822},"Post-confirmation training recovery",[2823,2824,2825,2826],"Preserve the poisoned v1 and v2 bounded-recovery journals, all 216 v2 inference rows, and their sealed candidate as immutable historical evidence. The failed confirmation grants no qualification, retention certification, or promotion authority and cannot be retried or reused.","Authorize a new ordinary development generation from the exact sealed lineage-139,015,512 candidate solely as an optimizer-preserving development initialization. The guardian remains the only trainer owner, and only future sealed descendants of this new recovery identity may be resumed.","Require the north-Pallet waypoint to observe Oak's actual event flag before advancing, and guide premature Lab entry back to Pallet only during that pre-starter phase.","Keep the existing nine-worker 3\u002F3\u002F3 starter cohort and append-only `runs\u002Ftrain` telemetry path so the singleton metadata uploader continues publishing the new execution and episodes to the consolidated Cloudflare D1 database.",{"version":2828,"date":2808,"sections":2829},"14.0.23",[2830,2837,2843],{"name":2831,"changes":2832},"Bounded recovery",[2833,2834,2835,2836],"Preserve the complete V10 qualification as immutable failed evidence. Every starter supplied 12\u002F12 scripted-cohort authority proofs and battle exposures; Bulbasaur recorded three wins and nine blackouts, while Charmander and Squirtle each recorded nine wins and three blackouts. Bulbasaur missed the unchanged four-win\u002Feight-blackout gates, so no V10 snapshot is valid.","Replace outcome-driven ancestor selection with one predeclared bounded development execution initialized from the failed V10 checkpoint at lineage 138,010,968. Run exactly 218 nine-worker PPO rollouts (1,004,544 optimizer transitions) to lineage 139,015,512 and permit exactly one candidate.","Fail closed on interruption, source drift, resume-after-start, a second execution, intermediate or replacement checkpoint selection, extension, or ancestor fallback. Preserve the exact optimizer state and conservative PPO settings while granting no ordinary training authority.","Reserve fresh confirmation seeds 100100001–100100036 for 36 stochastic attempts per starter. Require 36\u002F36 authority proofs and battle exposures, 36\u002F36 binary-resolved outcomes, at least 14 wins, and no more than 22 blackouts for every starter under the predeclared intersection-union gate. Failure is terminal; a pass still requires a deliberate new recovery-contract binding.",{"name":2838,"changes":2839},"Operations and reporting",[2840,2841,2842],"Teach the guardian-owned trainer path to seal one immutable pre-drain candidate at the exact optimizer boundary and poison any interrupted or drifting execution instead of resuming it.","Report bounded-development startup, active runtime binding, evaluation-lease waits, confirmation handoff, and terminal holds explicitly. Keep V10 visible as historical evidence without misclassifying its expected failure as a new live alert under the deliberate bounded contract.","Move production telemetry writes to a fresh D1 generation while preserving the capacity-bound predecessor database read-only, and keep aggregate spatial history exact across the cutover.",{"name":2844,"changes":2845},"Web dashboard",[2846,2847,2848,2849,2850,2851,2852],"Move the release changelog from the crowded run-atlas overview to a dedicated, prerendered page backed by a compact release-notes payload.","Expand run analysis into a zoomed-out, horizontally scrollable journey Sankey with ordered checkpoints, including Route 1 before Viridian City and an optional Route 22 branch.","Make each Sankey ribbon selectable and list its exact matching episodes below the chart, with direct access to the existing full run-detail view.","Limit visible-tab telemetry refreshes to one request cycle every five minutes while preserving an immediate initial load and a due refresh on return.","Replace full-table D1 episode cursors and version counts with indexed cursors and materialized totals, and publish atlas updates from an ingestion cursor.","Archive terminal episode telemetry after 30 days to checksum-verified, immutable R2 batches before bounded nightly D1 deletion.","Preserve the durable static\u002FR2 best-run winner when a hot telemetry shard has no stronger semantic candidate, so a shard cutover cannot regress the public best-run view.",{"version":2854,"date":2855,"sections":2856},"14.0.22","2026-08-18",[2857,2863],{"name":2858,"changes":2859},"New recovery generation",[2860,2861,2862],"Preserve the complete 72\u002F72-row v9 qualification as immutable failed evidence. Every starter supplied 12\u002F12 exact scripted-cohort proofs and battle exposures; Bulbasaur recorded three wins and nine blackouts, Charmander seven wins and five blackouts, and Squirtle nine wins and three blackouts. Bulbasaur alone missed the unchanged four-win\u002Feight-blackout gates, so no v9 snapshot is valid.","Reject the failed 3,999,996-step generation and every pre-existing descendant, then select the next eligible earlier root under the exact public-telemetry ranking fixed before v8: the sealed 14.0.16 checkpoint at execution step 2,999,997, lineage step 138,010,968, SHA-256 `b6b22afd185139e3882f7054c5db3de6dc668424f8f4862dbe74f76b42791c9f`.","Bind recovery `full-run-v3-semantic-stability-v9-2999997` to that root with contract SHA-256 `996d42817e2ea108da9082b075a81c2136ba715965ed3181b74c65558a0caff3`, preserving its structurally healthy Adam state and the conservative PPO stability settings.",{"name":2864,"changes":2865},"One-shot v10 qualification",[2866,2867],"Reserve fresh seeds 32101–32112 for a new inference-only v10 qualification, preserving the 12\u002F12 starter-proof and exposure requirements, four-win and eight-blackout gates, and nonblocking deterministic diagnostics.","Preserve v9 as a strict, snapshot-free, poison-on-interruption historical failure. V10 cannot import or resume any v7–v9 evidence, and only its passing sealed result may authorize its immutable non-training snapshot.",{"version":2869,"date":2855,"sections":2870},"14.0.21",[2871,2876],{"name":2858,"changes":2872},[2873,2874,2875],"Preserve the complete v8 qualification as immutable failed evidence: Bulbasaur supplied all 12 exact starter proofs and battle exposures but recorded only three wins and nine blackouts, missing the unchanged four-win and eight-blackout gates. No v8 snapshot is valid.","Reject the failed 4,999,995-step generation and every pre-existing descendant, then select the next eligible earlier root under the exact public-telemetry ranking fixed before v8: the sealed 14.0.16 checkpoint at execution step 3,999,996, lineage step 139,010,967, SHA-256 `4d4972f995b89c08b3b2071d5d94c0ea774fc23ada7f2a0845306dfa042a1771`.","Bind recovery `full-run-v3-semantic-stability-v8-3999996` to that root while preserving its complete finite Adam state and the existing conservative PPO stability settings.",{"name":2877,"changes":2878},"One-shot v9 qualification",[2879,2880],"Reserve fresh seeds 31101–31112 for a new inference-only v9 qualification, preserving the 12\u002F12 starter-proof and exposure requirements, four-win and eight-blackout gates, and nonblocking deterministic diagnostics.","Keep v8's strict poison-on-interruption and exact source\u002Fruntime binding; extend the narrow historical-failure verifier so sealed failed evidence remains authentic after a deliberate later recovery source change.",{"version":2882,"date":2855,"sections":2883},"14.0.20",[2884,2890],{"name":2858,"changes":2885},[2886,2887,2888,2889],"Preserve the completed v7 qualification as immutable failed evidence: the Squirtle cohort supplied only 10\u002F12 exact scripted-cohort starter proofs and 10\u002F12 battle exposures, so the otherwise satisfied win\u002Fblackout gates cannot authorize a snapshot or launch from the 11,999,988-step root.","Define recovery generation `full-run-v3-semantic-stability-v7-4999995` from the earlier sealed 14.0.16 periodic checkpoint at execution step 4,999,995, lineage step 140,010,966, and SHA-256 `10131272b8e3090bd856e56e8003d2e349751c0c34933820f65fc950693a8252`.","Select that root before v8 evaluation from healthy sealed periodic ancestors earlier than the failed v7 root, requiring Parcel delivery for every starter in the following production million and ranking exact starter-proof plus battle-exposure coverage before proof, opening-win, Parcel, aggregate coverage, and earlier-checkpoint tie breakers.","Record the selected root's public following-million evidence: exact proof and battle exposure were 14\u002F14, 8\u002F8, and 8\u002F8; opening wins were 6\u002F14, 5\u002F8, and 6\u002F8; and Parcel deliveries were 3\u002F14, 4\u002F8, and 3\u002F8 for Bulbasaur, Charmander, and Squirtle. These observations select the root but do not qualify it.",{"name":2891,"changes":2892},"One-shot v8 qualification",[2893,2894],"Reserve the fresh inference-only v8 panel at seeds 30101–30112 with the same predeclared four-win\u002Feight-blackout gates, complete authority proof and battle-exposure prerequisites, and nonblocking deterministic diagnostics.","Keep the guardian stopped until the exact v8 matrix seals successfully and its immutable snapshot exists; v7 evidence is never retried, completed rows cannot be resumed into v8, and no failed qualification can create a snapshot.",{"version":2896,"date":2855,"sections":2897},"14.0.19",[2898,2904],{"name":2899,"changes":2900},"Recovery qualification",[2901,2902,2903],"Define recovery generation `full-run-v3-semantic-stability-v6-11999988` from the sealed 14.0.16 checkpoint at execution step 11,999,988, lineage step 147,010,959, and SHA-256 `d4a85c108f352bf901b7b5b29cda3623ea144f2b3abd829c4cfeb5eea582f646`.","Select that single root before evaluation by maximizing the minimum per-starter opening trainer-win fraction in the following production million among pre-e9 periodic roots with nonzero Parcel delivery for every starter; preserve its complete, finite Adam optimizer state.","Add the immutable inference-only v7 qualification with twelve previously unused stochastic certifying seeds, deterministic nonblocking diagnostics, and proportionally scaled four-win\u002Feight-blackout per-starter gates. The historical 271xx and 281xx panels cannot be reused to choose a root, threshold, or retry.",{"name":2905,"changes":2906},"Evaluation and guardian clarity",[2907,2908,2909],"Keep PPO at nine balanced training workers and zero inline evaluation workers; expose frozen-checkpoint qualification as a separate external lane in the dashboard.","Preserve cryptographically verified failed qualification reasons as a fresh, exit-zero guardian hold even after source-only drift, while keeping missing, malformed, or tampered evidence as nonzero admission errors.","Forbid snapshot creation from a failed qualification and continue to block checkpoint selection and launch until the active root has both passing qualification evidence and its immutable snapshot.",{"version":2911,"date":2855,"sections":2912},"14.0.18",[2913,2921,2926],{"name":2914,"changes":2915},"Deliberate semantic recovery",[2916,2917,2918,2919,2920],"Define reviewed recovery candidate `full-run-v3-semantic-stability-v5-12999987` from the sealed 14.0.16 root at execution step 12,999,987, lineage step 148,010,958, and SHA-256 `e9a01b6ba6400dca3552feb29d1ba00d8a17b130d53b46c83f4c9bd6dbdd183d`.","Review the earlier root using its immediate production-million proxy: 13\u002F41 eligible episodes delivered the Parcel (Bulbasaur 4\u002F15, Charmander 5\u002F11, Squirtle 4\u002F15), compared with 3\u002F39 after the later 15,999,984-step root.","Preserve failed intermediate recovery `full-run-v3-semantic-stability-v4-15999984` and its v5 qualification as historical evidence: the root at execution step 15,999,984 and lineage step 151,010,955, with SHA-256 `acc890970cb7e16057bee25da6847c790c2952d09c73d5f53ead2f1a54ec402f` produced zero Bulbasaur wins and three blackouts in three attempts, failed qualification, and therefore never received a snapshot.","Reject that failed root and every 14.0.17 descendant, including the 999,999- and 1,999,998-step periodic checkpoints and the 2,744,757-step terminal checkpoint, instead of selecting a newer checkpoint from a regressed branch.","Clear the latched 14.0.17 semantic-quality hold only through this changed recovery contract; source-only releases and recovered rolling metrics remain unable to resume a rejected generation.",{"name":2922,"changes":2923},"Optimizer stability",[2924,2925],"Preserve the candidate root's structurally healthy Adam optimizer state, class, and keyword arguments; verified descendants preserve their archived optimizer state on subsequent resumes.","Reduce PPO updates to a `1e-5` learning rate, one epoch, `target_kl: 0.0025`, `max_grad_norm: 0.3`, and `ent_coef: 0.005` while retaining the recurrent policy, observation, and action contracts.",{"name":2927,"changes":2928},"Retention authority",[2929,2930,2931,2932],"Complete the v6 six-seed rows with stochastic wins of 1\u002F6, 5\u002F6, and 4\u002F6 and blackouts of 5\u002F6, 1\u002F6, and 2\u002F6 for Bulbasaur, Charmander, and Squirtle. Bulbasaur fails both the minimum-two-win and maximum-four-blackout gates, so no baseline snapshot is authorized.","Preserve the first completed v6 row set separately after final sealing rejected source drift caused by verifier code changing during evaluation; rerun the canonical frozen source to seal the same failed evidence.","Keep v4 and the failed, snapshot-free v5 qualification immutable as historical evidence with no authority to unlock the reviewed candidate.","Keep the current v5\u002Fe9 source contract review-only: v6 cannot authorize a launch, no snapshot may be created, and training remains held.",{"version":2934,"date":2855,"sections":2935},"14.0.17",[2936,2943],{"name":2937,"changes":2938},"Semantic recovery generation",[2939,2940,2941,2942],"Start recovery generation `full-run-v3-semantic-stability-v3-15999984` from the audited sealed 14.0.16 checkpoint at execution step 15,999,984 and lineage step 151,010,955 while preserving its optimizer state and PPO stability settings.","Exclude the regressed 18,713,088-step predecessor terminal tip from resume selection; only the exact new root and its cryptographically verified new- generation descendants are eligible.","Reserve `runs\u002Ffull_run_battle_retention\u002Fv4\u002Flatest` for fresh 14.0.17 retention evidence instead of inheriting the predecessor release's result.","Fail closed before installed-release checks, checkpoint selection, or launch until the public v4 verifier accepts the exact root's capability qualification and immutable non-training baseline snapshot.",{"name":2944,"changes":2945},"Training control",[2946,2947,2948,2949,2950,2951,2952,2953],"Make the file-invoked full-run guardian import its shared semantic reporter from the repository root, matching the production cron entrypoint.","Treat unavailable semantic-quality evidence as an explicit non-launching blocked state instead of reporting the trainer healthy, with subprocess and active-runtime regression coverage for the production import context.","Keep a statistically separated fixed-window Parcel-delivery collapse in the regressed state when one late delivery enters the rolling window, so the guardian cannot miss the checkpoint-safe latch between polling ticks.","Scope a semantic-quality hold to its recovery contract across source-only releases, preventing a version bump from making the same regressed branch tip selectable; only a deliberate new recovery generation clears the hold.","Preserve a committed semantic-quality handoff across release, config, and source refreshes so the exact trainer is never signalled twice or relabelled as an ordinary restart.","Keep a latched semantic regression visible while battle-retention owns the execution lease, and report that evaluation returns to hold without launch.","Move the status-facing 14.0.16 terminal-retention retry to a fresh immutable output after the first complete matrix correctly failed final sealing when its audited worktree identity changed during evaluation.","Treat `pokemon_fainted` as the required companion marker of a bounded whiteout in battle-retention evidence, while continuing to reject genuine non-blackout safety failures.",{"version":2955,"date":2956,"sections":2957},"14.0.16","2026-08-17",[2958],{"name":2959,"changes":2960},"Route 1 catch eligibility",[2961,2962,2963,2964],"Keep pre-Ball wild encounters out of the one-encounter-per-area ledger, then latch ordinary Nuzlocke spending when the first legal catch becomes possible.","Preserve a newly registered encounter's transient eligibility on its first policy frame so an immediate legal `THROW_POKEBALL` is not remapped to flee.","Revisit the audited Route 1 grass corridor immediately after the Mart's five-Ball gate, while leaving throw, battle, or flee entirely policy-owned.","Attest the catch unlock, post-Ball detour, and policy authority in guardian and status contracts, and report policy-selected catch macros as policy-owned.",{"version":2966,"date":2956,"sections":2967},"14.0.15",[2968],{"name":2969,"changes":2970},"Latched semantic-stability recovery",[2971,2972,2973,2974,2975],"Start a new recovery generation from the sealed 14.0.13 periodic checkpoint at step 7,999,992 instead of treating the later 14.0.14 descendant as independent clean-start retention evidence.","Preserve the checkpoint optimizer state while pinning resumed PPO updates to a `2e-5` learning rate, three epochs, `target_kl: 0.005`, and a `0.3` maximum gradient norm.","Add a target-independent, one-shot 1,000 reward for the last policy-owned Oak escort transition into the controller handoff anchor, and give waypoint 14 a local 24,000-step progress lease for the opening rival battle.","Make semantic regression or plateau request one checkpoint-safe guardian handoff and latch the same release\u002Frecovery generation in a stopped hold even if the rolling metric subsequently appears healthy.","Keep controller-created starter party, level, and catch accounting reward-neutral, including a delayed Pokédex-owned bit that previously could race the rebase and masquerade as the policy's first capture.",{"version":2977,"date":2956,"sections":2978},"14.0.14",[2979],{"name":2980,"changes":2981},"Full-run semantic recovery",[2982,2983,2984,2985],"Treat scripted starter initialization as reward-free controller bookkeeping, rebasing party, Pokédex, and level trackers without catch or lease credit.","Remove the controller-owned starter-match reward while retaining exact target verification, mismatch penalties, and invalid terminal behavior.","Apply an audited PPO target-KL early-stop guard to fresh and resumed recurrent policies to contain destructive optimizer updates.","Resume from the last sealed periodic checkpoint before the observed policy collapse, then admit only cryptographically verified descendants of that recovery root instead of drifting back to the rejected lineage tip.",{"version":2987,"date":2988,"sections":2989},"14.0.13","2026-08-16",[2990],{"name":2991,"changes":2992},"Battle-retention evaluation integrity",[2993,2994,2995,2996],"Reapply and attest each fixed evaluation seed after recurrent-checkpoint loading so candidate and baseline resets cannot silently inherit different saved training seeds.","Certify the policy's stochastic training-time behavior under a versioned retention contract while retaining deterministic inference as an explicit, non-certifying diagnostic.","Hold one typed execution lease for the complete evaluation and seal cycle, with atomic guardian arbitration that prevents an evaluator and trainer from acquiring emulators concurrently.","Keep verified retention outcomes as direct release-quality blockers without relabeling a fresh, exactly bound trainer as operationally unhealthy, and print the per-starter failure causes needed for remediation.",{"version":2998,"date":2988,"sections":2999},"14.0.12",[3000],{"name":3001,"changes":3002},"Audited Parcel-return guidance",[3003,3004,3005,3006,3007],"Replace broad off-phase first-visit rewards with bounded projected-exit shaping: only a new best distance or a one-shot audited transition pays, and lease renewal remains interval-gated and episode-persistent against farming.","Add the trace-verified Blue's House exit plus southbound Route 1 and Oak's Lab approach anchors while preserving uninterrupted bedroom-start policy runs.","Report a config-derived Parcel-return funnel, per-starter acquisition and delivery conversion, and fixed non-overlapping 256-episode Wilson windows.","Preserve complete pre-selection retention failures as failed matrix cells, and report zero-decision failed cohorts without misclassifying their sealed evidence as corrupt.","Keep battle penalties unchanged so this navigation release remains separable from the pending battle-retention certification.",{"version":3009,"date":2988,"sections":3010},"14.0.11",[3011],{"name":3012,"changes":3013},"Parcel-return learning recovery",[3014,3015,3016,3017],"Restore positive, first-visit navigation feedback on audited off-phase detours after broad suppression caused Parcel-return performance to collapse.","Keep the audited exit observations and Route 22 eight-badge frontier gate, and pin the corrected reward contract in guardian and status validation.","Add coupled regression coverage so production detour guidance cannot again coexist with a fully suppressed navigation signal.","Retain completed, legal, exact-cohort semantic bests as passive content-addressed checkpoints across release handoffs, without changing the guardian's maximum-lineage resume selection.",{"version":3019,"date":3020,"sections":3021},"14.0.10","2026-08-15",[3022],{"name":3023,"changes":3024},"Verified starter-phase alignment",[3025,3026,3027],"Align stale opening waypoint guidance to Oak's Lab exit only after the bounded cohort controller proves a coherent, exact assigned-starter match, without awarding skipped waypoint rewards or executing additional inputs.","Pin the reward-free synchronization marker in guardian and status contract validation so target allocation alone can never masquerade as observed selection evidence.","Retain same-release Parcel acquisition evidence beyond the rolling semantic window so an aged loss of acquisition performance remains a regression.",{"version":3029,"date":3030,"sections":3031},"14.0.9","2026-08-14",[3032],{"name":3033,"changes":3034},"Full-run semantic quality and status accuracy",[3035,3036,3037,3038,3039],"Give only the audited post-Parcel Oak's Lab return waypoints a longer progress lease, and suppress positive exploration credit on maps outside the pending ordered phase without hiding their detour telemetry.","Keep same-release Parcel deliveries in the semantic history so a later loss of delivery performance is reported as a regression instead of a missing milestone.","Report active vector episodes independently of clean promotion authority, separate semantic-quality blockers from operational liveness, and describe absent battle-retention evidence during an exact active run as pending the guardian-sealed terminal candidate.","Publish installed release and source identities on healthy guardian ticks so release status no longer falls back to `unreported`.","Bind battle-retention evaluation exclusion to the live training execution lease, fail closed when that lease disagrees with its pending manifest, and ignore abandoned unlocked manifests during terminal-candidate preparation.",{"version":3041,"date":3030,"sections":3042},"14.0.8",[3043],{"name":3044,"changes":3045},"Parcel-delivery navigation",[3046,3047],"Guide the policy through Oak's Lab along the collision-checked return aisle, then admit every historically demonstrated tile adjacent to Oak while still requiring the real Parcel-delivery transition for semantic completion.","Charge the checkpoint-compatible out-of-Mart shopping action as a policy-owned no-op so repeated idle selections receive the configured action penalties without changing the recurrent policy's ten-action interface.",{"version":3049,"date":3030,"sections":3050},"14.0.7",[3051],{"name":3052,"changes":3053},"Parcel-return phase alignment",[3054,3055],"Synchronize a real, policy-owned Parcel acquisition to the Mart-exit return phase when geometric waypoint credit is stale, without executing controller inputs or awarding skipped-route rewards.","Record the monotonic semantic alignment for audit and isolate its evidence from the plateaued 14.0.6 cohort.",{"version":3057,"date":3058,"sections":3059},"14.0.6","2026-08-13",[3060],{"name":3061,"changes":3062},"Parcel-return learning",[3063,3064],"Aim parcel-delivery guidance at Oak's audited interaction tile with exact distance shaping, while leaving the final facing and interaction inputs under uninterrupted policy control.","Isolate post-fix semantic evidence from the plateaued 14.0.5 cohort so a future delivery must be demonstrated by the corrected release itself.",{"version":3066,"date":3058,"sections":3067},"14.0.5",[3068],{"name":3069,"changes":3070},"Full-run training quality",[3071,3072,3073,3074],"Gate map-frontier reward by the active story phase and require eight badges before Route 22 can earn frontier credit, while retaining detours as telemetry.","Rank general checkpoints by ordered semantic waypoint progress before raw map priority so an accessible late-game side map cannot displace story progress.","Add audited off-phase exits for optional Viridian detours and region-aware Route 2 guidance back through Viridian Forest without rewinding the objective.","Report balanced current-release semantic progress windows and raise an explicit training-quality alert for regressions or sustained plateaus.",{"version":3076,"date":3077,"sections":3078},"14.0.4","2026-08-12",[3079,3086],{"name":3080,"changes":3081},"Opening-battle learning",[3082,3083,3084,3085],"Ignore transient, uninitialized opponent HP while establishing battle-damage baselines so real HP loss supplies the intended dense learning signal.","Stop penalizing directional battle-menu navigation, explicitly penalize rejected battle-item actions, strengthen bounded damage shaping, and reduce excess policy entropy while preserving policy-owned combat.","Remove the repeatable battle-entry bonus so controller-fled wild encounters cannot out-reward meaningful damage, trainer wins, or route progress.","Pin the corrected battle-learning settings in the full-run guardian and continue the recurrent v3 lineage from its gracefully sealed checkpoint.",{"name":3087,"changes":3088},"Certification integrity and observability",[3089,3090,3091,3092,3093,3094,3095,3096],"Require starter-target and selection-authority proof from every fixed-seed candidate and retained-baseline battle attempt, including losing cells.","Bind the retention evaluator's exact checkpoint, scenario, seed, inference, and job matrix back to the canonical bedroom-rival contract, and reject non-blackout safety failures from either side of the paired cohort.","Preserve fresh\u002Fresume launch provenance from the bound execution manifest in healthy guardian status, rejecting missing or contradictory resume evidence.","Fail closed in the independent status report when the configured or guardian-reported entropy, battle-reward, launch, or config identity drifts from the bound 14.0.4 runtime.","Rebuild the run-level furthest-map frontier from completed episode telemetry on resume so a graceful handoff cannot make reported progress move backward.","Restore the canonical total-party-level frontier across execution handoffs instead of reading the similarly named legacy health field.","Bound interrupted subprocess-vector teardown after a partially consumed rollout step so an atomically sealed guardian handoff cannot hang forever.","Let the guardian safely retry a terminal-teardown signal that failed before delivery, while rechecking process start time to prevent PID-reuse races.",{"version":3098,"date":3077,"sections":3099},"14.0.3",[3100,3106],{"name":3101,"changes":3102},"Starter controller hardening",[3103,3104,3105],"Finalize the transient starter award with cancel-safe inputs so timing differences cannot enter the nickname keyboard or strand an uninitialized party slot.","Guard both sides of the starter table before policy input and keep controller aborts out of cohort mismatch binding, reward, and error counters.","Require the durable followed-Oak story flags before the controller can own the Lab interaction, and guide fresh-bedroom workers through the north Pallet trigger before any Lab waypoint. A pre-Oak visit to the identical table tile now remains ordinary policy-owned exploration.",{"name":3107,"changes":3108},"Runtime and evidence integrity",[3109,3110,3111],"Bind guardian adoption, handoff, health, manifests, and execution leases to the exact Python source-tree digest, with a fail-closed legacy migration and highest-lineage same-run checkpoint resume selection.","Recompute battle-retention certification from the fully sealed plan, checkpoint inputs, fixed matrix results, and diagnostic evidence; outer summary edits or missing\u002Ftampered frozen inputs can no longer certify a release.","Bind status health to the guardian's exact PID, run, execution, config, and source identity before reporting operational or release readiness.",{"version":3113,"date":3077,"sections":3114},"14.0.2",[3115],{"name":3116,"changes":3117},"Guaranteed starter assignment",[3118,3119,3120],"Give Oak's Lab starter selection to a bounded, fail-closed cohort controller that uses ordinary game inputs, targets only the worker's assigned ball, and verifies the initialized party slot before returning control to the policy.","Record the pickup honestly as `scripted_cohort`, suppress sampled-action reward leakage during the macro, and terminate safely on any unexpected Lab state instead of permitting a wrong-ball fallback.","Protect the controller coordinates, timing bounds, authority, terminal contract, status metadata, and all three ROM pickup paths with regression tests.",{"version":3122,"date":3077,"sections":3123},"14.0.1",[3124,3129,3137],{"name":3125,"changes":3126},"Full-run starter cohorts",[3127,3128],"Keep each worker's Lab waypoint aligned to its assigned starter until the policy completes that pickup, instead of advancing target-agnostic guidance toward the exit while the party is still empty.","Defer starter-cohort scoring and Nuzlocke faint detection until the newly written party slot has coherent level and HP data, preventing transient starter-pickup frames from becoming false deaths or invalid attempts.",{"name":3130,"changes":3131},"Release operations",[3132,3133,3134,3135,3136],"Derive the full-run dashboard's no-heartbeat release fallback from the package version instead of pinning a release number in the component.","Gate every guardian launch on the installed distribution matching the source release, and bind graceful version handoffs to the adopted trainer's PID and execution identity so stale health cannot stop a newer process.","Report battle outcomes and rejected battle actions by starter, distinguish operational liveness from release readiness, and fail release evidence closed when retention artifacts are missing, stale, mismatched, or tampered.","Make the v14 full-run guardian and status reporter the documented operating path, and mark the former supervisor, specialist, canary, and hierarchical procedures as archived references that must not be revived.","Replace the README's stale scaffold goals and legacy June progress snapshot with the current v14 contract, required assets, safe verification commands, and an explicitly archived analyzer marker block.",{"name":3138,"changes":3139},"Battle retention evaluation",[3140,3141,3142],"Add a provenance-verified, inference-only full-run battle-retention lane for fixed all-starter bedroom-to-rival cohorts, including immutable baseline snapshots, paired win\u002Fblackout\u002Faction-rejection regression checks, a 40,000 step opening horizon, and a status-facing fail-closed summary with no training or promotion authority.","Require a same-run, earlier-lineage baseline before certifying retention; bind candidates to the sealed configured v3 latest model\u002Fconfig\u002Falgorithm\u002F observation identity, keep unpaired checks capability-only, and count battle wins relative to reset-time cumulative counters.","Preserve a digest-verified, content-addressed `best_battle` checkpoint lane independently for each starter, ranked by trainer wins before aggregate combat progress and published atomically without replacing navigation-owned `best_reward` selection.",{"version":3144,"date":3077,"sections":3145},"14.0.0",[3146,3153],{"name":3147,"changes":3148},"Continuous full-run lineage",[3149,3150,3151,3152],"Add `nuzlocke-full-run-recurrent-v2`, a fresh recurrent lineage whose every episode loads only the audited first-controllable bedroom state and then runs continuously to Hall of Fame, blackout, a meaningful-progress stall, or an immediate Nuzlocke-invalid terminal action. Disable promoted-state resets, progressive replay, emulator-state capture, and inline deterministic eval.","Preserve infrequent policy-weight checkpoints for crash recovery without resetting episodes, diversify canonical resets with seeded pre-control idle frames, and persist policy-owned starter selection plus RNG attestation in exact episode telemetry.","Add the incompatible `nuzlocke-full-run-recurrent-v3` lineage with an explicit nine-worker target pool: three workers each for Bulbasaur, Charmander, and Squirtle. Condition the shared policy on a target-starter one-hot, require policy-owned selection to match that target, and use a 768-sample PPO minibatch over each 4,608-step rollout.","Cut unattended ownership over to the dedicated nine-worker, zero-evaluator guardian and retire the specialist curriculum, clean-start canary, hierarchical readiness poller, specialist reviewer, and stuck checker.",{"name":3154,"changes":3155},"Sealed specialist predecessor",[3156],"Preserve the exhausted Viridian-to-Pewter continuation and its final checkpoints as historical evidence only. It no longer owns training, evaluation, promotion, or reset-state authority after the full-run cutover.",{"version":3158,"date":3159,"sections":3160},"13.0.0","2026-08-10",[3161,3166,3173,3179,3186,3192],{"name":3162,"changes":3163},"Automatic cohort qualification",[3164,3165],"Reserve immutable challenger, baseline, scenario, and state identities for every bounded Stage-10 or full-game milestone cohort, then run the paired semantic evaluator automatically after the trainers become idle.","Fail closed on missing, partial, stale, or tampered evaluation evidence and issue at most one exact continuation token only after the challenger is safety-clean and non-regressing. Evaluation evidence never promotes a model directly.",{"name":3167,"changes":3168},"Policy and specialist lineages",[3169,3170,3171,3172],"Make opening rehearsal genuinely clean-heavy with an adaptive 80%, 75%, or 70% clean-start reset share while retaining every required ladder rung in a spaced deterministic per-worker cycle and failing closed if a rung is absent.","Add matched fresh 73-feature feed-forward and recurrent PPO lineages with unit-normalized game state, plus recurrent hidden-state and episode-boundary handling across training, evaluation, canary, and hierarchical inference. Embed their ordered feature, normalization, lineage, algorithm, policy, and network contract inside each checkpoint and reject incompatible resumes.","Split specialist contracts into shared recurrent navigation and starter-scoped battle policies, require digest-bound trained promotion evidence for publication, and preflight the complete hierarchy before any emulator step. The monolithic end-to-end PPO remains benchmark-only.","Cut unattended training ownership over to a fail-closed specialist guardian: pin the reviewed shared-navigation inputs, resume only sealed exact-contract checkpoints toward one cumulative six-million-step target, bound unchanged launch retries, and keep the legacy monolithic supervisor disabled.",{"name":3174,"changes":3175},"Shared-navigation canary cutover",[3176,3177,3178],"Cut the scheduled singleton canary over from held starter-specific policies to the newest immutable, provenance-bound periodic checkpoint from the developmental shared recurrent Viridian-navigation lineage. Rotate all three clean starter states through the same upstream-trained navigation policy so newer canary snapshots reflect experience pooled during shared training; canary attempts themselves remain inference-only and never update one another or the trainer.","Preserve the source policy's Viridian-Mart objective, six-waypoint target set, 6,000-step horizon, and normalized 73-feature recurrent contract through the complete trained phase. Once the Mart milestone is reached, dynamically activate an evaluation-only route that acquires and delivers Oak's Parcel, exits the Lab, and continues under a 16,000-step meaningful-progress lease with a 500,000-step hard cap. Verified Center healing can renew the lease; merely occupying the Center cannot. Give the continuation a distinct series identity so historical Mart-terminal attempts cannot inflate its status or dashboard counts. Keep the entire canary allocation to one emulator outside the eight-worker plus one-periodic-evaluator training pool.","Label shared checkpoints as developmental and unpromoted, load no battle specialist, and grant no training, promotion, publication, hierarchical readiness, fallback, or production authority. Retain the prior starter-specific Boulder-Badge canary as a separately invoked manual legacy control whose results cannot be combined with the scheduled parcel-continuation series.",{"name":3180,"changes":3181},"Bounded learning semantics",[3182,3183,3184,3185],"Preserve Stage-10 ordered partial-progress returns instead of clawing every ordinary failed rehearsal back to the same total; retain a small fixed failure cost and the existing hard Nuzlocke safety penalties while reducing the nearly uniform policy's entropy pressure.","Settle verified Route-3 handoffs when the map byte changes even if no blank transition frame is observed, preventing stale Center-exit coordinates from handing the remaining Pewter corridor back to PPO.","Let Route-3 readiness recovery recognize exact-full-health as the sole remaining objective condition and permit a second verified, action-bounded post-heal return, while retaining the one-return default and a hard cap of two.","Reserve the Route-3 canonical worker for promotion-authority states and use appended Stage-10 candidates only as non-authority learning-lane fallbacks until stable progressive replay states exist.",{"name":3187,"changes":3188},"Reproducible training and evaluation telemetry",[3189,3190,3191],"Add immutable per-execution manifests with source, dependency, input, resume-lineage, failure, and checkpoint identities, plus append-only PPO update history.","Add exact per-episode behavior, resource, reward-component, milestone, safety, recovery, and terminal-outcome telemetry; keep sampled step traces out of automatic action-rate tuning and stop cumulative species inflation.","Add fixed, immutable candidate-versus-baseline semantic evaluation cohorts, retention matrices, and opt-in hash-bound curriculum promotion gates.",{"name":3193,"changes":3194},"Dashboard clarity",[3195],"Explain the clean-start canary's evaluator and checkpoint versions beside their reported values, distinguishing the code used for the test from the frozen model weights and why their release numbers can differ.",{"version":3197,"date":3198,"sections":3199},"12.1.64","2026-08-09",[3200],{"name":3201,"changes":3202},"Causal Route 3 readiness learning",[3203,3204,3205,3206,3207,3208],"Move the deterministic post-Brock Route 3 handoff into bounded reset preprocessing, preserve its semantic proof without PPO reward or clock credit, and permit only one verified post-Center rearm per episode.","Shape only stable, delta-based party readiness; renew bounded episodes only on durable party progress or confirmed wins, and prevent controller travel from manufacturing frontier, waypoint, or objective reward.","Assign the three learner workers distinct canonical, uniform-replay, and priority-replay roles; diversify recorded cohort seeds and tune resumed PPO exploration and critic weights for the readiness objective.","Record policy\u002Fcontroller\u002Frejected action ownership and require sufficient policy-owned data and authoritative episodes before promotion or automatic milestone renewal.","Retain compatible 12.1.63 Route 3 battle states across the source-only release while filtering replay to healthy Route 3 battle boundaries.","Load the bounded Route 3 learners on CPU because the current Torch runtime aborts while restoring the Stage-9 archive directly onto MPS.",{"version":3210,"date":3198,"sections":3211},"12.1.63",[3212],{"name":3213,"changes":3214},"Elite Four handoff integrity",[3215,3216,3217],"Add an explicit HM04 milestone after Surf: acquire Strength, teach move 70 to a living party member, and revalidate that living user before Victory Road can hand off to Indigo Plateau.","Require every newly written starter-track promotion, team signature, and safety total to come from the exact current release while preserving the separate historical-revalidation path.","Derive minimum-party-level progress only from a complete, positive live-party vector; incomplete vectors and unproven aggregate fields can no longer renew a cohort or authorize semantic progress.",{"version":3219,"date":3198,"sections":3220},"12.1.62",[3221],{"name":3222,"changes":3223},"Route 3 readiness and promotion integrity",[3224,3225,3226,3227,3228,3229],"Keep healthy post-Brock starts on a verified Gym\u002FCenter\u002FPewter handoff to Route 3, and rotate a safe under-level teammate into the lead before weak wild encounters can starve it of participation experience.","Route top-region Viridian Forest recovery north toward Pewter while retaining the southbound Viridian branch, covering the current Charmander poison and low-HP failure frontier without weakening recovery priority.","Include minimum party level in the clean durable milestone frontier and bind newly created promotion candidates to the exact current release, while preserving explicitly separate historical revalidation.","Correct the Silph Co frontier and specialist routing to start at canonical Silph Co 1F (map 181), not Saffron Mart (map 180).","Require Cut and Surf to be taught to a living party member before their HM milestones promote, retaining the party-wide proof in candidate sidecars.","Add code-owned recovery corridors from east Route 4 through Cerulean Center, and movement-only retreat routes for Nugget Bridge, Route 25, Bill's cottage, and Cerulean Gym.",{"version":3231,"date":3198,"sections":3232},"12.1.61",[3233],{"name":3234,"changes":3235},"Current-contract promotion authority",[3236],"Require starter-tracked curricula to earn their promotion successes under the current model contract, preventing historical rows from an obsolete objective definition from ending a rematerialized cohort on its first vector step.",{"version":3238,"date":3198,"sections":3239},"12.1.60",[3240],{"name":3241,"changes":3242},"Elite Four milestone throughput",[3243,3244,3245],"Publish the CLI-resolved worker count into every environment so a 3-worker cohort shards weighted static\u002Freplay cycles as three workers instead of the stale eight-worker YAML default.","Instantiate SB3's reward-only periodic evaluator only when enabled and turn it off for milestone stages, restoring the literal nine-learner plus one-canary emulator budget.","Make Route 3 an at-least-two-member handoff where every current party member is level 16 and fully healthy, with durable training replay; normalize Route 3\u002FMt. Moon battles through the bag-safe strongest-damaging-move controller while filtering stale off-route replay states.",{"version":3247,"date":3198,"sections":3248},"12.1.59",[3249],{"name":3250,"changes":3251},"Milestone replay integrity",[3252,3253,3254],"Balance Mt. Moon's authoritative Route 3 predecessor and progressive replay states evenly, preserving clean static-start promotion authority while giving learned deep starts equal rehearsal weight.","Credit ordered warp waypoints only from an exact, internally produced macro proof that matches the waypoint index, source tile, direction, and sampled destination; policy-supplied or mismatched evidence remains ineligible.","Keep replay-started episode provenance permanently promotion-ineligible, even if a later reset path looks canonical, and require clean semantic authority when recording a promotion candidate.",{"version":3256,"date":3198,"sections":3257},"12.1.58",[3258],{"name":3259,"changes":3260},"Mt. Moon forward progress",[3261,3262,3263,3264],"Replace Route 3's unreachable RAM-injected `(66, 0)` waypoint with the real collision-walked `(59, 0)` north connection and extend ordered guidance through Route 4, Mt. Moon 1F\u002FB1F\u002FB2F, the mandatory trainer and fossil gates, and the genuine east exit.","Replay only clean, current-release, stage-local waypoint\u002Fbattle\u002Fitem\u002Fheal captures while retaining the authoritative Route-3 start for promotion, and checkpoint twice within each bounded milestone cohort so useful learning survives exits.","Permit one level-cap-safe damaging finisher against a nearly defeated ordinary trainer when no healthy preventive switch exists; Gym, wild, unsafe, and repeated attempts remain fail-closed.","Attribute milestone progress to the clean static-start semantic frontier when progressive replay is enabled, keeping rehearsal states out of review gates.",{"version":3266,"date":3267,"sections":3268},"12.1.57","2026-08-08",[3269,3275],{"name":3270,"changes":3271},"Milestone frontier integrity",[3272,3273,3274],"Rank Route 4's reused west-side map before the Mt. Moon Center and cave floors, while retaining the exact east-exit coordinate as the completion proof. Entering the west cave approach can no longer pre-credit the final map frontier or its full bounded-cohort bonus.","Suspend ordered waypoint shaping only when deterministic recovery actually owns the action. A route-missing or out-of-range policy fallback can now approach the verified corridor and earn genuine semantic progress.","Include Pewter and its Pokemon Center in the Mt. Moon stage's required recovery-map preflight, so the complete Route 3 heal leg remains code-owned and fails closed if either verified map route regresses.",{"name":3276,"changes":3277},"Supervisor fail-closed scheduling",[3278,3279,3280],"Transfer the shared nine-environment pool only after every parsed process in the previous lane checkpoints and exits, derive availability and inventory from one process snapshot, and reserve legacy launches before spawning.","Validate either surviving supervisor ledger before scheduling, keep pending Stage-10 approval tokens bound to their starter, and give Stage-10 startups the same bounded fresh-health lease as milestone cohorts.","Prune review authority after a valid promotion, report checkpointed\u002Fheld lanes from actual active work, and tie fresh `running` health to its parsed live stage process before attributing the runtime release.",{"version":3282,"date":3267,"sections":3283},"12.1.56",[3284],{"name":3285,"changes":3286},"Bounded milestone review",[3287,3288,3289,3290,3291,3292,3293],"Reserve Route 3's west exit for verified recovery while the party is injured, preventing a healthy Mt. Moon cohort from immediately backtracking into Pewter from its Stage-9 handoff boundary.","Describe full-budget milestone review blocks as completed bounded cohorts in authoritative status instead of labeling them interrupted, keep explicitly authorized milestone trials out of the blocked set, and retain the Stage-10 semantic threshold while the milestone lane owns the supervisor status.","Replace a consumed Stage-10 or milestone approval with a fresh, one-use review block after the bounded worker completes, interrupts, disappears with running health, or exits without attributable current-release health.","Require the exact release\u002Fstarter\u002Fblock token for every Stage-10 or milestone trial, reject persistent bare starter IDs, and atomically reserve the token, launch attempt, and milestone startup lease before spawning a trainer.","Fail closed on malformed supervisor ledgers, reserve approvals for already active bounded cohorts before any early return, and reconcile a consumed Stage-10 result before returning the shared pool to milestone scheduling.","Preserve an exact Stage-10 approval across the complete milestone runner\u002Fworker checkpoint handoff, and select the newest matching-generation primary\u002Flast-known-good ledger in both the supervisor and status reporter.","Report exact current-release review-block tokens for both lanes without treating token availability as human authorization or a monitor threshold as every starter's hold trigger.",{"version":3295,"date":3296,"sections":3297},"12.1.55","2026-08-07",[3298],{"name":3299,"changes":3300},"Recovery-safe waypoint progress",[3301,3302],"Suspend ordered waypoint rewards, index advancement, and progress-lease renewal on every recovery-navigation action, so a westbound Route 3 heal cannot impersonate eastbound milestone progress.","Preserve the active semantic waypoint through the complete recovery route and resume ordinary advancement only after control returns to the policy.",{"version":3304,"date":3296,"sections":3305},"12.1.54",[3306],{"name":3307,"changes":3308},"Route 3 progress guidance",[3309,3310,3311],"Close the bounded Charmander cohort's Route 3 recovery gap at `(19, 4)` and replay that exact current-release battle-end state through a clean, battle-free Pewter Center heal.","Add ordered eastbound Route 3 waypoints plus post-heal Center and Pewter exit guidance while preserving the active semantic waypoint across recovery.","Distinguish the genuine east Route 4 cave exit from Route 4's west entrance: Mt. Moon completion now requires B2F entry and the post-cave `(24, 6)` tile.",{"version":3313,"date":3296,"sections":3314},"12.1.53",[3315],{"name":3316,"changes":3317},"Route 3 recovery closure",[3318,3319],"Extend the code-owned Route 3 recovery profile from the west transition to the north, middle, and south interior corridors observed in the bounded 12.1.52 milestone cohort, then retreat through Pewter to its registered Pokemon Center without opening field-item menus. Keep trainer-flag variants with no battle-free westbound path outside the authorized trial scope.","Cover every 12.1.52 `route_out_of_range` terminal coordinate with a verified branch and replay every Charmander Route 3 battle-end state, plus representative live states from the other map components, through a battle-free Center heal before authorizing its bounded milestone trial.",{"version":3321,"date":3296,"sections":3322},"12.1.52",[3323],{"name":3324,"changes":3325},"Milestone cohort safety",[3326,3327,3328,3329,3330],"Read the Gen-I bag from its authoritative item count and `$FF` terminator, failing closed on malformed slots instead of retaining sentinel or stale storage as phantom inventory and item-transition evidence.","Block ordinary `A` confirmation on the trainer-battle ITEM command whenever battle items are disabled, preserving the existing ten-action checkpoint interface while preventing raw menu navigation from bypassing Nuzlocke item rules. Keep legal wild-catch macros and the RUN column unchanged.","Preserve navigation-first recovery when a weakened party is outside a verified corridor: return control to the policy under the bounded route-gap guard instead of immediately opening the less reliable field-healing menu.","Report battle-item and invalid-catch episode endings in milestone failure frontiers so semantic review cannot mistake those violations for an empty failure surface.","Preserve a fresh one-use milestone approval through prelaunch processing; the terminal health that created its review block now becomes the next cohort baseline instead of replacing the approved block before launch.",{"version":3332,"date":3333,"sections":3334},"12.1.51","2026-08-06",[3335,3340,3346,3351],{"name":3336,"changes":3337},"Verified Mt. Moon recovery",[3338,3339],"Close the Mt. Moon recovery preflight with ROM-replayed routes for all three cave floors and Route 4's west approach to the registered Center. Select disconnected B1F\u002FB2F components independently and require Route 4 map 15 as part of the milestone's complete recovery chain.","Re-select a persisted recovery branch after a same-map warp coordinate settles, while preserving ordinary one-tile movement and two-tile ledges. Exhaustive component checks and genuine low-HP and poisoned-menu PyBoy replays now cover the cave-to-Center retreat without trainer engagement.",{"name":3341,"changes":3342},"Canary runtime and telemetry",[3343,3344,3345],"Resolve clean-start checkpoints from the held clean-opening lane instead of the milestone supervisor's top-level stage, so starters 1, 4, and 7 evaluate their Stage-10 checkpoints while the downstream lane is independently held.","Publish live canary status and run heartbeats with exact completed, error, interrupted, and incompatible counts, checkpoint provenance, coordinates, and distinct attempt outcomes. Keep the reviewed 16,000-step progress lease unchanged while labeling lease and step-limit endings as bounded recycling.","Atomically claim cloud outbox envelopes, preserve producer replacements, suppress superseded mutable run state, and identify uploader release\u002FPID so the guardian can perform a verified two-tick release handoff.",{"name":3347,"changes":3348},"Web canary status",[3349,3350],"Expose the latest canary heartbeat separately from training metadata and add a dashboard card for evaluator release, checkpoint model, freshness, current attempt, proof counts, and bounded-recycle outcomes.","Recover legacy endpoint coordinates from nested final locations, retain completed canary rows that are not spatially mappable, and keep the public endpoint-kind contract limited to objective, violation, or other.",{"name":3352,"changes":3353},"Orchestration safety",[3354],"Preserve milestone semantic-review blocks across release and configuration changes. A held starter now requires a one-use, release-scoped `POKEMON_MILESTONE_TRIAL_STARTERS` approval, and every approved bounded cohort returns to review instead of authorizing another launch.",{"version":3356,"date":3357,"sections":3358},"12.1.50","2026-08-05",[3359],{"name":3360,"changes":3361},"Guarded Route 3 correction",[3362,3363,3364],"Record the live 12.1.49 Route 3 failure mode: low-HP post-Brock starts could enter the field-Potion macro before their verified Pewter recovery route, and bounded menu cleanup could abort the attempt with movement still unsafe.","Make navigation-first recovery a code-owned profile invariant, so the real Bulbasaur, Charmander, and Squirtle Stage-9 starts follow their verified Gym and Pewter Center corridors before field healing is considered.","Define the corrected-release resume contract: the first 12.1.50 Route 3 cohort selects the validated immediate Stage-9 predecessor instead of any interrupted 12.1.49 same-stage checkpoint. Tracked milestone checkpoints are eligible only when health proves both their release and model provenance match the current source release; unknown or prior-release checkpoints fall back to the exact validated predecessor without deleting their artifacts.",{"version":3366,"date":3357,"sections":3367},"12.1.49",[3368,3374],{"name":3369,"changes":3370},"Guarded post-Brock transfer",[3371,3372,3373],"Pin every downstream starter track to the audited Stage-9 PPO interface: twelve ordered memory features over four frames, the same ten ordered actions and transition semantics, and a rollout-compatible batch size of 512. Reject an incompatible resume before creating any PyBoy worker and isolate ROM\u002FRAM worker namespaces by starter and stage.","Require one candidate-backed, starter-bound model\u002Fstate\u002Fpromotion triple from the exact preceding milestone. Revalidate party identities and HP, encounter\u002Fcatch\u002Fdeath ledgers, ruleset, release, transition evidence, and promotion authority before skip, routing, specialist publication, or the next stage; final consolidation also requires the complete prior chain.","Extend in-episode map, item, badge, Gym identity\u002Froster, and level-cap proof through every configured milestone instead of accepting pre-satisfied endpoint states.",{"name":3375,"changes":3376},"Recovery and supervisor fail-closed guards",[3377,3378,3379],"Add PyBoy-verified recovery from all three real post-Brock Gym states and from Route 3's west entrance back through Pewter to the Center. Materialized stages now declare required recovery maps; Mt. Moon, Bill, and Misty remain explicitly blocked until each missing route is emulator-verified.","Account for every live trainer globally, checkpoint duplicate runners or workers and mixed\u002Foversubscribed lanes, stop consolidation when its clean gate closes, and persist startup failures or ten-minute no-health timeouts so a release cannot crash-loop.","Scope Stage-10 approval to the exact durable semantic block, require a current-release fully finished bounded cohort before milestone renewal, and keep prior-release health from becoming a new baseline.",{"version":3381,"date":3357,"sections":3382},"12.1.48",[3383,3389],{"name":3384,"changes":3385},"Post-Brock milestone lane",[3386,3387,3388],"Decouple audited post-Brock milestone training from the strict clean-opening rehearsal gate, preserving Stage 10 as independent release and final consolidation evidence while starter-specific tracks advance in stage-major lockstep toward Misty and later milestones.","Seed the new versioned full-game tracks only from validated paired Stage-9 artifacts, reserve promotion authority for the immediate predecessor state, and retain older states as training-only rehearsal.","Bound each milestone launch to one rollout-aligned cohort below 500K steps, share the fixed nine-environment pool without oversubscription, and hold a non-advancing durable frontier for semantic review.",{"name":3390,"changes":3391},"Transition and safety evidence",[3392,3393,3394],"Record episode-start badges, map and item transitions, badge deltas, and recognized Gym Leader entry evidence in promotion sidecars and episode telemetry so pre-satisfied replay states cannot manufacture progress.","Require Misty promotion to prove a legal one-to-two badge transition after a recognized level-21-cap battle, and carry the verified Pewter recovery route into the first post-Brock stage.","Report clean-opening and downstream milestone lanes independently, including their active, held, and next-action state.",{"version":3396,"date":3357,"sections":3397},"12.1.47",[3398,3404,3409],{"name":3399,"changes":3400},"Stage-10 reporting clarity",[3401,3402,3403],"Report completed `progress_stalled` attempts as bounded progress-lease recycling under monitoring, not as trainer-stuck alerts unless the supervisor has emitted an explicit current semantic-review block.","Show active clean semantic progress separately as provisional liveness that cannot satisfy the durable opening gate until a valid clean episode ends.","Distinguish prior Stage-9 Brock proof from current Stage-10 canary successes, legal opening rehearsals, and distinct-team requirements in authoritative and scheduled status language.",{"name":3405,"changes":3406},"Stage-10 semantic phase control",[3407,3408],"Defer forced wild-battle readiness training until the audited first-capture handoff, so Route 1 encounters before waypoint 28 are fled instead of consuming the bounded opening cohort on premature starter grinding.","Preserve legal catch precedence at the handoff and fail closed when the live semantic waypoint cannot be interpreted.",{"name":3410,"changes":3411},"Bounded flee recovery",[3412,3413],"Abort an attempt safely after three consecutive flee-menu synchronization timeouts in one wild battle, rather than allowing a bad command or cursor state to consume the remainder of its progress lease.","Report exhausted flee synchronization as its own stable environmental ending and cover command-menu, cursor-normalization, and outcome-timeout paths.",{"version":3415,"date":3416,"sections":3417},"12.1.46","2026-08-04",[3418],{"name":3419,"changes":3420},"Bounded cohort integrity",[3421,3422,3423,3424,3425],"Stop treating a live PID with terminal or mixed-release health as a healthy member of the starter pool; give a newly launched worker one observation to publish running health, then drain a persistently finalized worker so its allocation can be relaunched on the current release, while status reports the bounded recovery or release handoff as an expected stop.","Preserve each live Stage-10 worker's environment width when another starter exits, leaving freed slots idle instead of restarting survivors with another full budget; hold an interrupted current-release cohort for semantic review rather than automatically granting it another 497,664 steps.","Fail closed on current-release running health with no live worker and on an already oversubscribed Stage-10 pool, checkpointing the latter instead of accepting more than the fixed nine-environment budget.","Fit Stage 10's per-episode hard limit inside each worker's share of the fixed 497,664-step cohort, guaranteeing at least one completed episode per worker even when one starter receives all nine environments.","Keep the semantic plateau baseline strictly durable. A valid clean episode already active at the deadline receives one fair completion opportunity, while invalidation, terminal health, or a post-deadline retry cannot renew or repeatedly defer the review boundary.",{"version":3427,"date":3416,"sections":3428},"12.1.45",[3429],{"name":3430,"changes":3431},"Clean-frontier liveness",[3432,3433,3434],"Publish a provisional semantic frontier only from active, rule-valid, promotion-authority clean episodes, allowing the aggregate Stage-10 plateau monitor to observe real progress before a vectorized episode finishes.","Keep provisional progress out of promotion and opening-gate evidence while using it for supervisor and hourly-monitor liveness, so replay workers still cannot advance or reset the clean semantic contract.","Persist clean map ID and progression priority for completed and in-flight clean episodes, and include map priority in plateau comparisons and status.",{"version":3436,"date":3416,"sections":3437},"12.1.44",[3438],{"name":3439,"changes":3440},"Final semantic safeguards",[3441,3442,3443,3444,3445],"Treat a field heal as successful only after both the Potion effect and menu closure are verified; abort to a safe checkpoint if bounded cleanup leaves a menu open, before recovery navigation can emit directional input.","Distinguish Route 2's disconnected southern and northern regions even though they share map ID 13, routing a regressed northern-phase attempt back through Viridian Forest without changing its semantic waypoint index.","Match the clean-start canary's policy-visible episode-progress normalization to Stage 10's 30,000-step horizon, avoiding a canary-only observation shift.","Honor an explicitly supplied callback release throughout clean-frontier restoration and health, summary, and run-stat persistence.","Label a stopped trainer as an intentional semantic hold or release handoff only when the supervisor record is fresh, current-release, and explicit.",{"version":3447,"date":3416,"sections":3448},"12.1.43",[3449,3457],{"name":3450,"changes":3451},"Stage-10 semantic recovery",[3452,3453,3454,3455,3456],"Reset the meaningful-progress distance baseline after each ordered waypoint, and use monotonic four-tile approach credits so a completed landmark cannot suppress legitimate progress toward the next one.","Project waypoint features onto the latest phase-consistent transition when a policy regresses to another map, with explicit exits for Route 22 and optional Viridian interiors instead of comparing unrelated map coordinates.","Bound failed field-Potion attempts, fall back immediately to verified Center navigation, and apply a retry cooldown so an unavailable Start menu cannot monopolize the remainder of an episode.","Route low-HP Viridian Mart replay starts through the stable south exit before continuing along the existing Viridian-to-Center recovery corridor.","Start the repaired release from a fresh 497,664-step budget: exactly 108 complete nine-environment PPO rollouts, below the reviewed 500K ceiling.",{"name":3458,"changes":3459},"Evaluation and supervision integrity",[3460,3461,3462,3463],"Persist a per-release semantic frontier assembled only from completed, rule-valid promotion-authority episodes, and use it for Stage-10 supervision, status, and stuckness checks so replay-ladder maxima cannot reset the gate.","Match the clean-start canary to Stage 10's 12K floor, 16K progress lease, four-tile distance interval, and story renewals while retaining its 60K observation scale and 500K absolute safety cap.","Exclude exhausted health from extension counts unless it belongs to the active release, preventing old terminal cohorts from consuming a new release's bounded retry allowance.","Report a trainer intentionally stopped for semantic review or release handoff as held instead of raising a false dead-screen runtime alarm.",{"version":3465,"date":3466,"sections":3467},"12.1.41","2026-08-03",[3468],{"name":3469,"changes":3470},"Stage-10 semantic frontier",[3471,3472,3473,3474,3475],"Hold the Parcel, delivery, and Poké Ball waypoints until their underlying story transitions actually complete, preserving route guidance through the Mart and Oak's Lab interactions.","Renew bounded episodes on story progress and allow a 16K-step progress gap, covering the measured clean Pallet-to-Lab frontier while keeping the current-release Charmander trial below its aggregate 500K ceiling.","Rank promotion-authority clean checkpoints by ordered waypoint progress before incidental map discovery, so detours cannot displace a stronger full-opening route checkpoint.","Record Parcel delivery once in the canary milestone buffer instead of repeating the durable flag on every subsequent step.","Distinguish a selected active bounded trial from the starters still held for semantic review in the authoritative next-action report.",{"version":3477,"date":3466,"sections":3478},"12.1.40",[3479,3484],{"name":3450,"changes":3480},[3481,3482,3483],"Restore the audited Mart, Pallet, Route 2, and Viridian Forest navigation frontiers to the full-opening route, and add the missing reach-Mart replay rung so the shared Viridian failure is practiced directly.","Preserve the first worker-sharded replay slots by leaving the initial reset to the vector environment, interleave clean starts throughout the weighted ladder, and recycle non-progressing openings within the bounded cohort.","Separate Brock-engagement shaping from the badge-confirmed terminal frontier, and classify battle wins, losses, and escapes from the game's battle-result byte instead of an opponent-HP proxy when available.",{"name":3485,"changes":3486},"Evaluation integrity",[3487,3488],"Save the best promotion-authority clean-start checkpoint and evaluate that exact immutable checkpoint cohort, preventing stale or changing checkpoint failures from stopping a new model.","Scope clean-attempt telemetry to the current release, normalize team compositions independent of party order, and retain faint evidence across the complete episode for trustworthy promotion review.",{"version":3490,"date":3466,"sections":3491},"12.1.39",[3492,3498],{"name":3493,"changes":3494},"Stage-10 training integrity",[3495,3496,3497],"Replace random and progressive Stage-10 rebasing with a deterministic, worker-sharded ladder cycle that audits every fixed phase once and reserves twelve canonical clean starts for the ten-rehearsal promotion gate.","Start each new bounded release from its starter's promoted Stage-9 model, while preserving current-release checkpoint resume after interruption, and cap the catch\u002Ftraining roster at three members for attainable team diversity.","Keep Brock's final waypoint active until a trainer battle actually begins, prevent off-phase map discoveries from renewing the progress lease, and normalize ordinary stalled returns without weakening Nuzlocke penalties.",{"name":3485,"changes":3499},[3500,3501],"Keep held Stage-10 tracks authoritative for clean-start canary selection and scope canary summaries to Stage 10, starter, evaluator release, and frozen model release.","Apply the ten-attempt canary stop gate per starter so one branch cannot stop or clear another, and report bounded evaluation holds without mislabeling them as zero-step semantic plateaus.",{"version":3503,"date":3466,"sections":3504},"12.1.38",[3505,3510],{"name":3506,"changes":3507},"Recovery reliability",[3508,3509],"Stop the synchronized Center-PC opener on the first actionable storage-menu frame, preventing an extra confirmation from selecting an empty Withdraw box before the macro can choose Deposit.","Cover the corrected transition with both a deterministic menu-boundary test and a real PyBoy replay of the captured post-faint Pewter recovery state.",{"name":3511,"changes":3512},"Supervision integrity",[3513,3514],"Hold a finished Stage-10 cohort for evaluation only after it reaches the configured training budget, so an unexpected under-budget clean exit remains eligible for supervised continuation.","Quarantine the bounded 12.1.37 cohort after its telemetry exposed the PC menu defect, retaining its atomic terminal checkpoints for diagnosis only.",{"version":3516,"date":3466,"sections":3517},"12.1.37",[3518],{"name":3519,"changes":3520},"Execution provenance",[3521,3522],"Create and rotate run\u002Fexecution identity for local-only training as well as cloud-backed runs, so every terminal checkpoint manifest is attributable to one concrete execution without enabling external uploads.","Start the final bounded cohort from the unchanged Stage-9 checkpoints after an emulator-backed 12.1.36 shutdown smoke verified both atomic model ZIPs; the identity-less smoke artifacts remain ineligible for evaluation.",{"version":3524,"date":3466,"sections":3525},"12.1.36",[3526,3531],{"name":3527,"changes":3528},"Checkpoint durability",[3529,3530],"Install the shutdown guard before emulator or model startup, immediately unwind blocked vector-environment reads on `SIGINT` or `SIGTERM`, and defer repeated signals until checkpoint commit finishes.","Stage and validate both `latest.zip` archives before atomic replacement, then publish terminal health with their exact paths, sizes, timestamps, and SHA-256 identities as the final execution commit.",{"name":3485,"changes":3532},[3533,3534],"Require the paired evaluator's fresh model to exactly match the current terminal checkpoint manifest and reject missing, stale, active, malformed, or identity-mismatched executions before creating evaluation output.","Start a fresh release-scoped cohort from the unchanged Stage-9 checkpoints; the short multi-execution 12.1.35 artifacts remain quarantined.",{"version":3536,"date":3466,"sections":3537},"12.1.35",[3538,3542],{"name":3506,"changes":3539},[3540,3541],"Route low-HP Oak's Lab replay states through a collision-verified Pallet and Route 1 corridor to Viridian Center, and bound failed PC deposit macros so a broken menu interaction cannot pin a training worker indefinitely.","Save the current model and terminal health after either `SIGINT` or `SIGTERM`, preserving a trustworthy checkpoint when the supervisor stops a stale or plateaued cohort.",{"name":3543,"changes":3544},"Evaluation and supervision",[3545,3546,3547],"Revalidate existing Stage-10 promotions against the current static-start, release, candidate-sidecar, and distinct-team contract before advancing.","Hold completed bounded Stage-10 cohorts for evaluation instead of silently relaunching them, and require a stable terminal current-release checkpoint with no active trainer before the paired evaluator can freeze it.","Start a fresh release-scoped cohort so interrupted 12.1.34 checkpoints and failure evidence cannot satisfy the repaired evaluation or promotion gates.",{"version":6,"date":3466,"sections":3549},[3550],{"name":3506,"changes":3551},[3552,3553,3554],"Advance only framebuffer-verified post-battle dialogue before recovery movement, without allowing generic collision handling to start NPC text.","Route standard Pokemon Center PC traffic through the collision-verified row-5 aisle in both directions, avoiding the blocked row-7 furniture.","Start a fresh release-scoped cohort so failed recovery evidence from the earlier routes cannot satisfy replay, evaluation, or promotion gates.",{"version":7,"date":3466,"sections":3556},[3557],{"name":3558,"changes":3559},"Reward integrity",[3560,3561],"Clamp every failed or invalid Stage-10 episode's complete return below zero, even when accumulated route and Gym rewards exceed its terminal penalty.","Start a fresh release-scoped cohort so checkpoints and promotion evidence produced before the cumulative-return fix cannot satisfy the repaired gate.",{"version":8,"date":3466,"sections":3563},[3564,3570],{"name":3565,"changes":3566},"Training recovery",[3567,3568,3569],"Prevent Route 1 recovery from starting fresh NPC dialogue, verify both failure hot spots with real PyBoy save states, and cover the complete low-HP trip to a successful Viridian Center heal.","Replace the monolithic opening rehearsal with a bounded, late-heavy state ladder, explicit Stage-10 waypoint rebasing, Gym-interior guidance, and readiness-aligned catch\u002Ftrain\u002Fflee controls for each starter.","Remove repeatable battle, level, flee, and healing reward proxies, quarantine older replay captures, and reserve promotion authority for canonical clean Route 1 starts.",{"name":3571,"changes":3572},"Monitoring",[3573],"Checkpoint and hold Stage-10 tracks after a bounded semantic plateau or a current-release 0\u002F30 clean-start canary, while reporting parcel, inventory, party, catch, waypoint, badge, and legal-success frontiers.",{"version":9,"date":3466,"sections":3575},[3576,3582],{"name":3577,"changes":3578},"Training reliability",[3579,3580,3581],"Keep the advanced level cap active between a verified Gym Leader defeat and the delayed badge-memory update, so legal Brock experience cannot terminate an otherwise successful attempt.","Synchronize ordinary and forced battle switches against exact menu RAM, accept verified direct send-outs, and retry one dropped party-menu opening.","Concentrate Boulder Badge replay on clean post-Junior-Trainer heals, avoid fallback attacks when any party member has reached the badge cap, and expand stuck Route 1 recovery through prompt clearing and the full detour sequence.",{"name":3571,"changes":3583},[3584],"Restrict clean-opening Brock stall alerts to opening-rehearsal canary attempts, while leaving general canary status and failure summaries intact.",{"version":10,"date":3586,"sections":3587},"2026-08-02",[3588,3591],{"name":3577,"changes":3589},[3590],"Replace the Viridian south-entry recovery shortcut rejected by live smoke testing with the collision-verified corridor around the Pokemon Center wall.",{"name":3571,"changes":3592},[3593],"Bind Hall-of-Fame success to the configured terminal canary target and infer it from legacy records only when eight-badge evidence is present.",{"version":11,"date":3586,"sections":3595},[3596,3601],{"name":3577,"changes":3597},[3598,3599,3600],"Let a solo under-level starter keep ordinary battle control instead of entering a return-to-finisher rotation with no finisher, and add the calibrated Route 1 retreat plus a south-entry Viridian Center branch required by low-HP recovery.","Guide Boulder training from the Pewter Gym entrance through the Junior Trainer to Brock, flee irrelevant wild battles, and reject replay snapshots above the legal first-Gym entry cap.","Prefer the supervisor's live active-stage assignment over stale historical progress baselines when selecting clean-start canary checkpoints.",{"name":3571,"changes":3602},[3603],"Report clean-opening Brock and Hall-of-Fame canary success separately, drive the opening-stall alert from the Brock milestone, and retain structured recovery and rotation details on terminal attempts.",{"version":12,"date":3586,"sections":3605},[3606,3610],{"name":3577,"changes":3607},[3608,3609],"Require every party to enter a Gym Leader battle at or below the current badge-indexed level cap, then activate the next Gym cap during the identified Leader battle so experience earned against the Leader remains legal.","When a threatened active Pokémon has no healthy switch target during an identified Gym Leader battle, return control to the policy while preserving faint retirement, forced switching, whiteout, and promotion legality checks.",{"name":3571,"changes":3611},[3612,3613],"Alert on repeated current-release Brock failures, the final automatic budget extension, and a rolling zero-success clean-start canary instead of treating raw step and episode movement as sufficient progress.","Refresh the Gym Leader level-cap policy from the authoritative Nuzlocke template when materializing starter-only Pewter and Boulder stages.",{"version":14,"date":3615,"sections":3616},"2026-08-01",[3617,3621],{"name":3577,"changes":3618},[3619,3620],"Track trainer victories separately from wild wins, require a legal Center heal after the Pewter Junior Trainer, and replay stable fully healed post-trainer checkpoints so later episodes can concentrate on Brock.","Add calibrated Pewter Gym exit branches and a collision-safe city-to-Center corridor, accept both verified nurse-facing coordinates, and bound missing, stuck, or non-progressing recovery routes with a safe checkpoint restart.",{"name":3622,"changes":3623},"Telemetry and dashboard",[3624,3625],"Preserve terminal wild and trainer battle context—including full trainer rosters—and publish it through episode logs and clean-start canary results.","Show each run's endpoint tile, terminal battle type, and opponent party in the atlas run inspector, while retaining useful party HP in canary details.",{"version":15,"date":3615,"sections":3627},[3628],{"name":3577,"changes":3629},[3630],"Reset stage-local waypoint state when entering Pewter recovery, the Brock battle, or a clean opening rehearsal, and prevent those stages from replaying snapshots whose waypoint indexes belong to a different objective.",{"version":16,"date":3615,"sections":3632},[3633,3640],{"name":3634,"changes":3635},"Evaluation",[3636,3637,3638,3639],"Give clean-start canary attempts a rolling progress lease beyond the original 60,000-step opening horizon, retain a 500,000-step hard cap, and continue after Brock with the existing full-game map progression toward Hall of Fame.","Preserve the checkpoint's calibrated pre-Brock waypoint observations and report Boulder Badge and progress-lease milestones separately from the final full-game canary objective.","Lock the badge-indexed level-cap schedule to each next Pokemon Red Gym Leader's highest-level Pokemon and cover all eight Gym transitions.","Report evaluator release and frozen-checkpoint model provenance separately, marking ambiguous legacy attempts as unreported instead of relabeling old weights.",{"name":1102,"changes":3641},[3642,3643,3644],"Publish completed clean-start canary evaluations with explicit canary labels, provenance, and filtering, without mixing their sampled movement into the training heatmap.","Render every observed Pewter interior with tile-aligned map art, connected placement, and authoritative bounds that discard map-transition coordinates.","Drive Nuxt map rendering and asset staging from the generated background registry so newly mapped areas cannot silently lose their artwork on the web.",{"version":17,"date":3615,"sections":3646},[3647],{"name":3648,"changes":3649},"Recovery integrity",[3650,3651,3652,3653],"Compose map-keyed and legacy-list Center additions over the verified standard registry, preserving future-city overworld routes during profile application.","Make registered layout routes authoritative inside Centers, reject malformed runtime routes safely, and prevent item macros from targeting retired party members even when stale game bytes show restored HP or status.","Allow config-only materialization to refresh future stages before predecessor promotion, so dormant direct-resume configs retain the current recovery rules.","Evaluate readiness and badge gates across the current surviving roster rather than requiring the original party size after a mandatory death deposit.",{"version":18,"date":3615,"sections":3655},[3656],{"name":3657,"changes":3658},"Nuzlocke integrity",[3659,3660,3661],"Make survivor continuation a runtime invariant even for stale materialized configs: a single faint is retired and recovery-gated, while only a full party whiteout is terminal.","Apply one shared survivor-recovery profile to generic curriculum stages, starter tracks, and Brock races so future materialization cannot silently disable forced switching, PC deposit, healing, or verified route recovery.","Validate normalized Center profiles, targets, and interior route contracts at environment startup; registered layouts are authoritative and uncalibrated custom interiors fail closed instead of inheriting standard coordinates.",{"version":19,"date":3615,"sections":3663},[3664],{"name":3657,"changes":3665},[3666,3667,3668],"Treat an individual faint as permanent retirement instead of ending the run, force a living replacement, and require exact-identity PC deposit before ordinary progress or checkpoint promotion resumes. Only a full-party whiteout now ends an attempt for fainting.","Synchronize RUN against the verified battle command menu and Gen I result markers, rechecking party safety after a denied escape so transitional inputs cannot accidentally choose FIGHT or sacrifice a weakened lead.","Normalize all eleven verified standard Pokémon Centers through one reusable layout profile with entrance-to-PC and post-deposit PC-to-nurse routing; custom interiors must declare calibrated profiles explicitly.",{"version":20,"date":3615,"sections":3670},[3671,3676],{"name":947,"changes":3672},[3673,3674,3675],"Keep under-ready reach-Pewter replay states in the verified Forest training corridor, rank party-floor gains ahead of route depth, and renew the progress lease when a catch or level gain advances the roster.","Return a switch-trained party member to a healthy finisher before attacking, synchronize both persistent Gen I party-menu cursors from RAM, suppress failed rotation loops for the encounter, and bound failed recovery with a safe escape instead of spending the episode on status moves.","Gate the Forest exit and Pewter Gym entrance on the configured roster, readiness, and recovery requirements; Brock remains a separate stage after a clean Pewter Center handoff.",{"name":3677,"changes":3678},"Reliability",[3679,3680],"Yield recovery navigation to the policy outside verified corridors, ignore status-only moves when evaluating battle PP, scope the Pewter recovery gate to the Gym door, and remove the route lease from recovery and Brock stages.","Align Boulder Badge promotion on one clean legal win across the trainer, supervisor, and status paths.",{"version":3682,"date":3683,"sections":3684},"12.1.19","2026-07-31",[3685,3689],{"name":3686,"changes":3687},"Orchestration",[3688],"Report fresh health-backed starter allocations when process enumeration is unavailable, and fail closed on stale unverified health rather than claiming a dead trainer is active or risking a duplicate launch.",{"name":947,"changes":3690},[3691],"Safely train a newly caught Forest party member to the existing party-wide readiness floor during reach-Pewter replay, then resume fleeing optional encounters once the full party is ready.",{"version":21,"date":3683,"sections":3693},[3694,3697,3701],{"name":3686,"changes":3695},[3696],"Add a five-minute cron guardian for the inference-only clean-start canary, with singleton process checks, current-release heartbeat verification, clean release handoff, and canary release state in the regular status report.",{"name":947,"changes":3698},[3699,3700],"Reorder the party from the overworld when its lead approaches the current badge level cap, choosing the weakest safe replacement and verifying the exact Pokémon identity swap before continuing.","Keep starter level, move, promotion, and replay checks independent of party order so a deliberately rotated lead remains eligible for the curriculum.",{"name":3702,"changes":3703},"Telemetry",[3704,3705],"Record a catch on the first sampled action after reset while excluding starter pickups, gifts, and PC withdrawals from catch-location events.","Count verified field lead rotations and failures, force each attempt into the step trace, and expose its cap, source, target, and identity details in run health telemetry.",{"version":3707,"date":3683,"sections":3708},"12.1.17",[3709,3714,3717,3720],{"name":3710,"changes":3711},"Training integrity",[3712,3713],"Preserve the encounter species and location through delayed and full-party capture confirmation, preventing a later field read from losing or misattributing the catch.","Require the Boulder Badge in opening-rehearsal states so downstream runs cannot satisfy the rehearsal gate from a pre-Brock snapshot.",{"name":3634,"changes":3715},[3716],"Add a frozen-checkpoint clean-start canary that rotates across all starter branches without training or promotion authority and reports generalization progress separately from advanced-state curriculum results.",{"name":3702,"changes":3718},[3719],"Record each confirmed capture as its own episode event with species, encounter area, tile coordinates, and step so capture origins stay accurate even when the episode ends elsewhere or a full party sends the Pokémon to the PC.",{"name":3721,"changes":3722},"Web atlas",[3723,3724],"Add the episode quilt and Kanto ghost map for exploring run outcomes and route coverage across the starter branches.","Show the caught area and tile in the Agent Pokédex and run details, while clearly labeling older captures whose exact location was never recorded.",{"version":22,"date":3726,"sections":3727},"2026-07-30",[3728],{"name":3686,"changes":3729},[3730],"Refresh the shared Stage 5 model\u002Fstate pair from a deterministic validated current-ruleset starter branch once all three capture candidates are present. This prevents a legacy aggregate artifact from crashing every battle-readiness launch before emulator environments are allocated.",{"version":23,"date":3726,"sections":3732},[3733,3741],{"name":3657,"changes":3734},[3735,3736,3737,3738,3739,3740],"Register an encounter area atomically before any catch, flee, switch, or attack macro, enforce Oak's Parcel at both the action and inventory layers, and preserve first-frame catches without misclassifying the caught species as a duplicate.","Retire blanket Viridian Forest ledger reopening. New runs catch the first legal Forest encounter normally, while pre-12.1.15 replay captures and promotions remain on disk but are excluded from new ruleset runs.","Reject ambiguous or illegal catch-training snapshots, including pre-Parcel, no-Ball, already-spent, fainted, and incompletely described encounters.","Match the strict web ruleset across curricula and the Brock race: any faint ends the attempt, and promotion cannot carry a death ledger forward.","Isolate Brock-race training captures by starter and stamp replay sidecars and promotions with the release\u002Fruleset version used to create them.","Scope supervisor relaunch and extension limits to the active release\u002Fruleset so exhausted historical checkpoints cannot block required legal retraining.",{"name":3721,"changes":3742},[3743],"Clarify that an empty model-filtered Pokédex means that release recorded no new catch event; inherited party ledgers are not misattributed as new catches.",{"version":3745,"date":3726,"sections":3746},"12.1.14",[3747],{"name":947,"changes":3748},[3749,3750],"Catch the first eligible non-duplicate Viridian Forest encounter during reach-Pewter navigation, while continuing to flee duplicates and already spent encounters.","Reopen Forest encounter ledgers polluted by the earlier forced-flee policy until a third species is caught, and retain three-member parties in progressive replay.",{"version":3752,"date":3726,"sections":3753},"12.1.13",[3754],{"name":947,"changes":3755},[3756,3757,3758],"Defer replay-state capture until menus, dialogue, fades, and battle transitions have settled; sanitize inherited reset states and exclude legacy transitional captures from progressive replay.","Restore the verified Gen I PKMN command path for safety switches.","Prevent recovery detours from immediately reversing route progress, and cancel replayed overlays before retrying Pokémon Center positioning.",{"version":3760,"date":3726,"sections":3761},"12.1.12",[3762],{"name":947,"changes":3763},[3764],"Give reach-Pewter attempts both a 12,000-step initial floor and rolling meaningful-progress renewal window, allowing longer Viridian Forest traversals.",{"version":3766,"date":3726,"sections":3767},"12.1.11",[3768],{"name":947,"changes":3769},[3770,3771,3772,3773],"Correct the Gen I safety-switch command path to open PKMN instead of ITEM, and verify the selected party member actually became the battle mon.","Make reach-Pewter workers flee optional wild encounters before safety switching, preventing avoidable faints and pre-Brock level-cap violations.","Replay clean, level-capped route frontiers by ordered-waypoint progress so Forest navigation can continue from earned deep states instead of restarting from Route 1 every episode.","Add a southbound Viridian Forest recovery corridor for weakened or poisoned parties returning to the Viridian Pokémon Center.",{"version":3775,"date":3726,"sections":3776},"12.1.10",[3777,3782],{"name":947,"changes":3778},[3779,3780,3781],"Resume reach-Pewter navigation from the Viridian waypoint after a center heal, preventing Route 2 and Forest checkpoints from being replayed in the wrong map.","Constrain non-objective battle controls to a bag-safe attack macro through the Brock and opening-rehearsal stages, while preserving legal catch macros.","Treat battles that finish during a safety-switch macro as safely resolved instead of terminating the episode as a failed switch.",{"name":3702,"changes":3783},[3784,3785,3786],"Separate center-positioning retries from real nurse-dialogue heal attempts and failures.","Publish current-release environmental termination counters and reset runtime safety counters across releases so historical incidents no longer appear as active regressions.","Clarify party-level and legal-objective labels in authoritative status.",{"version":3788,"date":3789,"sections":3790},"12.1.9","2026-07-29",[3791,3796],{"name":947,"changes":3792},[3793,3794,3795],"Replace the fixed 30,000-step reach-Pewter horizon with a progress-aware lease: meaningful route progress renews the attempt up to a 60,000-step hard cap, while workers that stop advancing are recycled after 5,000 stale steps.","Keep stall-penalty clocks separate from true progress clocks so periodic penalties can no longer prevent configured early termination.","Add a deterministic Route 22-to-Viridian recovery handoff and ordered waypoints so Charmander's legal readiness promotion can rejoin the Route 2 path instead of wandering off-route.",{"name":3702,"changes":3797},[3798],"Record progress-lease expirations as `progress_stalled`, including the last meaningful milestone, rolling deadline, and stale duration.",{"version":3800,"date":3789,"sections":3801},"12.1.8",[3802,3805],{"name":947,"changes":3803},[3804],"Correct the Gen I battle-menu ITEM slot guard so spent-route and duplicate encounters cannot consume Poké Balls through raw menu inputs, while legal first non-duplicate encounters remain catchable.",{"name":3702,"changes":3806},[3807],"Break environmental episode endings into stable reason codes for field safety, failed battle switches, early aborts, exploration stalls, battle limits, blackouts, and unknown terminal conditions; retain concise diagnostic details and cumulative counts in episode, health, summary, atlas, supervisor, and scheduled status data.",{"version":3809,"date":3789,"sections":3810},"12.1.7",[3811,3816],{"name":947,"changes":3812},[3813,3814,3815],"Extend battle-readiness episodes from 30,000 to 40,000 steps so party-level objectives have enough time to complete.","Let untreated overworld poison resolve as a faint instead of preemptively aborting the episode, while continuing to use an Antidote when available.","End battle-readiness episodes on the first faint under Nuzlocke permadeath so the policy receives the configured faint penalty and a clear terminal signal.",{"name":1102,"changes":3817},[3818,3819,3820,3821,3822,3823,3824,3825,3826,3827],"Migrate the generated run dashboard to a component-based Nuxt 4 application while retaining Tailwind, Cloudflare Workers static-asset deployment, and a production Docker image.","Move build and deployment ownership to Cloudflare Workers Builds and remove the duplicate GitHub Actions deployment.","Make the scheduled atlas refresh a direct JSON-only command with schema validation before it commits and triggers the Nuxt deployment workflow.","Stop tracking local runtime logs, ROM data, raw walkthrough imports, and generated Python package metadata; compact the published atlas JSON.","Select the newest recorded model version when the web atlas first loads, while keeping an “All model versions” scope available.","Let the endpoint inspector switch from a focused marker back to all endpoints and temporarily show attempts from every model version without changing the current model scope for the rest of the atlas.","Add cursor-centered wheel, trackpad, pinch, double-click, and keyboard zoom to the connected world atlas while preserving drag-to-pan navigation.","Open any listed attempt in an accessible run-details modal, including runs referenced by endpoint results and Pokédex catch history.","Publish this version changelog in the web atlas.","Replace the route-by-route world overview with one connected, zoomable atlas and add missing Route 22 and Viridian Forest backdrops.",{"version":3829,"date":3789,"sections":3830},"12.1.6",[3831,3835],{"name":947,"changes":3832},[3833,3834],"Synchronized field-Potion use to Pokémon Red's live Start, bag, item-action, and party cursors; verified exactly one Potion is consumed and the menu returns to the overworld.","Recorded deterministic low-HP return routes from Route 2 and Viridian City to the Viridian Pokémon Center, including exact nurse positioning and full-party heal verification.",{"name":3702,"changes":3836},[3837],"Added a dedicated healing event stream and persistent counters for Center attempts and failures, item heals, Potions consumed, and recovery-route starts and arrivals.",{"version":3839,"date":3789,"sections":3840},"12.1.5",[3841,3846],{"name":947,"changes":3842},[3843,3844,3845],"Added low-HP and poison recovery navigation, legal field healing, and the Route 1 Potion gift to the battle-readiness route.","Hardened forced-switch handling and battle-menu synchronization so a fainted active party member does not strand an otherwise legal attempt.","Added release-aware supervisor handoffs and checkpoint-safe restarts for battle-readiness workers that stop making measurable progress.",{"name":3702,"changes":3847},[3848],"Reported recovery, progressive replay, healing, and supervisor handoff state in run health and status output.",{"version":3850,"date":3789,"sections":3851},"12.1.4",[3852],{"name":947,"changes":3853},[3854,3855],"Corrected battle-readiness waypoint restoration so Center-started attempts resume the intended route instead of inheriting a completed route position.","Restarted the configured waypoint cycle after a successful Pokémon Center heal.",{"version":3857,"date":3789,"sections":3858},"12.1.3",[3859],{"name":947,"changes":3860},[3861,3862],"Added progressive replay from validated intermediate battle-readiness states to expose the policy to more of the Route 2 and Forest approach.","Rejected replay states that were unsafe, unadvanced, or incompatible with the active party and Nuzlocke constraints.",{"version":3864,"date":3789,"sections":3865},"12.1.2",[3866],{"name":947,"changes":3867},[3868,3869],"Made PPO minibatches compatible with the supervisor’s 3, 4, 5, and 9 environment allocations.","Preserved checkpoint\u002Fresume continuity as the shared nine-environment pool was redistributed between starter tracks.",{"version":3871,"date":3789,"sections":3872},"12.1.1",[3873],{"name":3686,"changes":3874},[3875,3876],"Tightened per-starter process ownership, checkpoint handoffs, and stage-major lockstep for Bulbasaur, Charmander, and Squirtle.","Improved promotion and resume bookkeeping for repeated battle-readiness extensions.",{"version":3878,"date":3879,"sections":3880},"12.1.0","2026-07-28",[3881,3885],{"name":3686,"changes":3882},[3883,3884],"Added the starter-track supervisor, structured health reporting, and Brock race tooling.","Added stale-observation recovery and safer starter lifecycle management.",{"name":947,"changes":3886},[3887],"Continued each starter independently through battle readiness while preserving validated curriculum promotions.",{"version":3889,"date":3879,"sections":3890},"12.0.1",[3891,3895],{"name":947,"changes":3892},[3893,3894],"Added per-starter curriculum runners and a clean promotion path through the Viridian story stages.","Expanded progression-safety coverage for supervisor, environment, reward, logging, and release behavior.",{"name":1102,"changes":3896},[3897],"Added richer map backdrops and improved run-atlas episode telemetry.",1792773948192]