Crystal/docs/SESSION_HANDOFF.md

Recovered Claude session

Latest continuation: see Controlled capture comparisons: 2026-09-30 at the end of this file. Earlier sections are historical snapshots.

Recovered on 2026-09-23 from the local Claude Code transcript ~/.claude/projects/-Users-kawa-Crystal/4ac88f30-29e6-4fd5-a4b2-b6a318dd83b3.jsonl. The session stopped at its usage limit at 09:54 UTC, during research, without a final handoff.

Resumed increment: completed 2026-09-23

The movement increment below is complete. The remaining sections preserve the original recovery context.

• Added directional two-tile ledge jumps with the existing arc/shadow animation. Unsupported terrain, invalid destinations and occupied landings remain refused.
• Corrected directional mats to wait for exit input. Warp arrivals keep their actual coordinates and appropriate facing; doors, stairs and caves animate a downward exit. Removing the old inward teleport also corrected Elm's scripted introduction position.
• Mid-walk, mid-jump and mid-connection saves retain the source position. Landing scenes rewind to the actual source of a jump, rather than the intervening wall.
• Restored compatibility with early version-3 saves lacking the money field.
• Replaced the scratch diagnostic with synthetic movement regressions and optional-ROM tests covering five Route 29 ledge types. Route 29 is enabled only inside that test fixture, not in production.
• Fixed the battle test suite's eager ROM read so missing-ROM runs really skip private tests. Updated browser stair inputs and added an exit-mat save/reload flow. Updated the roadmap and architecture to distinguish the existing battle code from its unfinished integration.

Final checks: 75 unit tests passed with the local ROM; without it, 55 passed and 20 private tests skipped. All 24 Chromium/WebKit browser tests passed. Native and reference builds, release isolation, pinned-reference checks, vendor checks and git diff --check passed. No commit or push was made.

Next: validate the overworld-to-battle browser flow and complete Route 29's remaining script, capture/tutorial and item interactions before enabling the route. Player turn timing also remains a documented comparison gap.

Scope and current state

The original request was to read ARCHITECTURE.md, PLAN.md and PRESENTATION_PLAN.md and continue the native port. The working tree contains substantial uncommitted work from multiple increments; preserve it when continuing.

• Opening scenes, scripted movement, starter selection, naming, the aide's Potion and version 3 saves are implemented and documented.
• Native battle modules, ROM tables, damage/type parity tests, encounter handling, battle UI and whiteout handling have been added. These later changes are not fully reflected in the architecture and roadmap documents.
• Only the original six maps are enabled in NATIVE_MAPS. Route 29 (24:3) remains disabled. Passing battle unit tests does not establish a playable Route 29 encounter or the full opening quest.

Exact stopping point

Claude was preparing to revise player movement in src/game/world/world.ts. Its last investigation concerned spawn facing after warps. It had not implemented the planned ledge/warp revision.

The pinned local PRET checkout contains the relevant routines:

• engine/overworld/player_movement.asm: .TryStep, .TryJump, .CheckLandPerms, .CheckWarp and the direction tables.
• engine/overworld/tile_events.asm: CheckWarpCollision and CheckDirectionalWarp.
• engine/overworld/player_object.asm: SpawnPlayer.
• engine/overworld/map_objects.asm: movement/turn timing.

Research found that ledge tops have land permission; the jump check uses the current tile and facing after an ordinary step fails. Doors use collision $71, downward warp mats $70, and stairs $7a. Directional warp mats require their designated input rather than an immediate warp simply upon entry. Verify these rules against the local routines before changing the existing movement flow.

tests/zz-scratch.test.ts is Claude's unfinished diagnostic: it logs local-ROM warp and ledge collisions, contains no assertions and unconditionally requires the private ROM. Replace it with meaningful optional-ROM or synthetic tests before treating the suite as ready for public CI.

Validation after recovery

Run against the current working tree with the local ROM and reference checkout:

• npm test: 45 passed, including the assertion-free scratch diagnostic.
• npm run build: passed.
• npm run test:e2e: 22 passed across Chromium and WebKit, including native damage/type comparisons against original routines in the reference emulator.
• npm run check:reference: ROM, pinned rebuild and profile addresses match.
• npm run check:release: native output excludes emulator and game assets.
• npm run check:vendor: passed.

These checks do not establish complete battle behavior, browser battle UI flows, Route 29 progression or physical-device compatibility. No gameplay source was changed during session recovery.

Original next-increment notes (movement now completed)

Finish ledge and directional-warp movement with focused regression tests, including interrupted movement/save behavior and preservation of the existing opening scenes. Audit Route 29 callbacks, scripts and unsupported destinations before enabling it. Then test wild encounters and battle controls through the browser, including win, escape, loss and save/reload, and reconcile the roadmap with the verified behavior.

Battle integration increment: 2026-09-23 (resumed from Codex)

Codex committed and pushed c08d250, then stopped mid-increment with the keyboard/D-pad battle test failing. Claude Code completed it:

• Browser battle flows run through the real UI via tests/battle-fixture.ts (encounter selection only; production maps stay gated): win with level-up and Smokescreen, escape, whiteout, keyboard/D-pad menus, mid-battle backup restore and unsupported-effect rollback. tests/battle-world.test.ts covers the same at world level, plus pending stat experience and whiteout callback rollback.
• Saves accept stored stats between the no-stat-exp and full-stat-exp values, since GiveExperiencePoints awards stat experience without CalcMonStats.
• A script warp (e.g. whiteout) validates its destination and commits healing, money loss and destination callbacks together; aborts restore state and map.
• Keyboard and on-screen D-pad navigate battle menus (battle-selected); A continues messages, B leaves the move list.
• Fixed a phone-width layout bug: a long notice sat in the Play button's auto grid column and crushed the status text, pushing the game panel off-screen. Below 720px the notice now has its own row.
• Test fixes: the navigation test now presses A through the intro message, and uses a real pointer click on the D-pad (synthetic pointer events cannot be captured).

Final checks: 82 unit tests with the local ROM (55 passed, 27 skipped without it); 36 Chromium/WebKit browser tests; native and reference builds; release, vendor and reference checks; git diff --check. Not committed.

Next: Route 29 (24:3) scripts, catching tutorial and items before enabling it; then capture, switching and PACK in battle.

Route 29 and Cherrygrove increment: 2026-09-23

Enabled Route 29, its Route 46 gate, Cherrygrove City and its five buildings. Item balls, the Pokémon Center nurse and the Poké Mart were each ported by a Sonnet sub-agent in an isolated worktree, reviewed against the disassembly and merged; the rest was done directly.

• Fruit trees run FruitTreeScript; a native CheckDailyResetTimer clears the daily engine flags once per game day. VAR_BOXSPACE reads an empty box.
• Item balls run FindItemInBallScript with a native TryReceiveItem.
• The nurse runs PokecenterNurseScript; HealParty now applies ComputeMaxPP's PP Up clamp. EnterMapWarp's spawn recording makes whiteouts return to the last Pokémon Center entered.
• Standard marts run natively (StandardMart/BuyMenu/SellMenu).
• Battle command recoil (Struggle) was the only refusal found by a local stress run of 360 Route 29 battles across starters and times of day.
• Save format 3 gained fruitTrees, dailyReset and lastSpawnMap (migrated by absence).
• The phone-width status bar layout and parity test map counts were fixed.

Still refused: catchtutorial and Cherrygrove's rival scene (both after the Mystery Egg), Route 30/46. NPCs do not roam yet.

Next: NPC roaming movement (map_objects.asm random walk/spin, radius and visibility rules, deterministic RNG in tests), trainer sight lines and trainer battles, Route 30 and Mr. Pokémon, capture and the catching tutorial.

Trainer AI continuation: 2026-09-24

The trainer foundation was subsequently committed through 3e92936. The next AI increment translates Opportunist, Aggressive and Risky scoring, including the ROM's stall/reckless/risky tables and Aggressive/Risky's deterministic pre-variation damage estimate. Smart is now the only refused move-scoring layer, so the rival remains gated. Continue with Smart before the rival, or with trainer sight lines and normal trainer-object scripts for the already-supported Youngster and Bug Catcher classes on Route 30.

AI layer review: 2026-09-24 (resumed from Codex)

Claude Code checked Codex's uncommitted Opportunist, Aggressive and Risky layers against engine/battle/ai/scoring.asm. The HP thresholds, AI_50_50 (≥128), 79 percent - 1 (200), last-wins ties, the power < 2 and reckless exclusions, the strict-borrow KO test and the constant-damage paths all match.

One fix: AIDamageCalc goes through EnemyAttackDamage and BattleCommand_DamageCalc, and both read wCriticalHit. A flag left over from an earlier move therefore selects unboosted stats (ignoring screens) and doubles the estimate. enemyAIDamage now reads the native criticalHit instead of hard-coding false. It also reuses truncate. A local-ROM test fixes the estimate at 4, 3 (Reflect) and 6 (leftover critical).

Known side effects of AIDamageCalc that are not modelled: it overwrites wCurDamage, wTypeModifier, wAttackMissed and hBattleTurn. CheckTurn resets the modifier and miss flag before each move. Whether the wCurDamage overwrite can reach Counter/Mirror Coat still needs checking before those effects are ported.

Final checks: 147 unit tests with the local ROM; 42 Chromium/WebKit browser tests; native and reference builds; reference, release and vendor checks; git diff --check. Not committed.

Next (unchanged): Smart AI before the rival, or trainer sight lines and normal trainer-object scripts for Route 30's Youngster and Bug Catcher.

Route 30 trainers: 2026-09-24

Chose option 2 from the review above: trainer sight lines and trainer-object scripts. Both Route 30 trainers have two Pokémon, so the increment also ports the trainer's send-out and switch AI. Route 30 and its Berry house are now enabled in production.

• World: _CheckTrainerBattle at step end (before warps, coordinate events and encounters) and on idle ticks; TalkToTrainer for OBJECTTYPE_TRAINER; BGEVENT_ITEM hidden items. A refused sighting is not restarted until the player moves (native safeguard against a rollback loop).
• Interpreter: loadtemptrainer, trainertext, trainerflagaction, scripttalkafter, endifjustbattled, checkjustbattled, writeobjectxy, encountermusic, loadmem wRunningTrainerBattleScript, readmem wHiddenItemID, and callasm TrainerWalkToPlayer/SetMemEvent. applymovementlasttalked wMovementBuffer decodes the native buffer.
• Battle: multi-Pokémon trainer parties (HandleEnemyMonFaint, EnemyPartyMonEntrance, FindMonInOTPartyToSwitchIntoBattle with its BattleRandom fallback), AI_SwitchOrTryItem/CheckAbleToSwitch/AI_Switch with the original type-matchup, e-carry and quarter-HP bugs, and wPlayerUsedMoves. BattleTrainer now requires switchFlags.
• Stress (local, not committed): 1,080 Route 30 battles; the only refusal was Chikorita's Reflect at level 12 (screen), a pre-existing gap.

Final checks: 160 unit tests with the local ROM; 44 Chromium/WebKit browser tests including the Route 30 sighting through the UI; native and reference builds; reference, release and vendor checks; git diff --check. Not committed.

Next: Mr. Pokémon's house (Oak, Pokédex, Mystery Egg) and the return to Elm; Reflect/Light Screen; then Smart AI before the rival. Joey (after the Egg) needs askforphonenumber and the phone standard scripts.

Reflect and Light Screen: 2026-09-24

Completed the battle gap found in the Route 30 stress run. screen now runs both moves from ROM effect scripts. Independent five-turn counters include the setup turn, do not refresh on repeated use, survive enemy send-outs and expire with ROM text in HandleScreens order. The existing damage path handles physical/special defense and critical stage rules.

Seven new local-ROM regressions cover player and enemy use, PP on failure, concurrent durations, expiration order, send-out persistence and damage stats. Also moved the NPC test's eager ROM read into beforeAll, restoring its intended skip behavior when no ROM is available.

Checks: 167 unit tests with the local ROM; 76 passed and 91 skipped without it; 44 Chromium/WebKit browser tests; native and reference builds; reference, release and vendor checks; git diff --check. Screen routines were reviewed against pinned PRET source, not compared by executing the original routines. Changes are uncommitted.

Next: Mr. Pokémon's house and the Mystery Egg; Smart AI before enabling the rival; capture and the catching tutorial. Joey's post-Egg phone scripts remain unsupported.

Mr. Pokémon and the return to Elm: 2026-09-24 (resumed from Codex)

Codex had enabled Mr. Pokémon's house and added NameRival, blackoutmod, reloadmap, the queued Elm call and the no-handler Smart AI subset. It stopped with all four quest browser tests failing and without docs. Claude Code finished the increment:

• The rival scene needed special FadeOutMusic (audio only, now accepted).
• The quest fixture's constant () => 0 RNG hung BattleCommand_DamageVariation, which redraws until the byte is at least 217. The fixture now uses a seeded LCG. Relocating to Cherrygrove had also carried the house's moved Mr. Pokémon and Oak onto the Guide Gent and rival (ids 2 and 3), producing an unrestorable save. The fixture now clears objects and starts off the Guide Gent's tile.
• The browser test expected specialPhoneCall 1; SPECIALCALL_ROBBED is 2.
• NameRival yielded a naming choice with zero pages, because promptbutton had already flushed them. The dialogue never reached its last page, so the officer scene soft-locked. It now shows the ROM's "RIVAL'S NAME?" and reads the "SILVER" default from ROM. As in InitName, blank names of any number of spaces take the default. The naming input's accessible label no longer says "Nickname".
• Elm's call now runs the ROM's CheckSpecialPhoneCall.script → Script_ReceivePhoneCall wrapper. It reads the condition and caller script from SpecialPhoneCallList instead of hard-coding id 2, adds memcall, and shows HangUp's timed "Click!". Ringing and the caller name box are still not presented.
• Added tests: the rival scene for all three starters and both triggers, a CANLOSE loss, reload after each scene, the Elm call (indoors, outdoors, mid-call reload), the officer naming (named, blank, all-space, reload at the naming screen), the Egg handover and the aide's Poké Balls, plus a browser naming-field test.

Checks: 191 unit tests with the local ROM (80 passed, 111 skipped without it); 50 Chromium/WebKit browser tests; native and reference builds; reference, release and vendor checks; git diff --check. Not committed.

Next: capture and the Route 29 catching tutorial (catchtutorial), the first remaining refusal in the opening quest. Then the ringing presentation, player switching and items in battle. Joey's post-Egg phone scripts remain unsupported.

Capture and the catching tutorial: 2026-09-24

Continued the roadmap with the first refusal remaining in the opening quest.

• Battle PACK opens the Balls pocket and runs PokeBallEffect: the Poké, Great, Ultra and Master Ball multipliers, the original HP arithmetic (including the overflow above 341 maximum HP), the sleep/freeze bonus (BRN/PSN/PAR add nothing), HELD_CATCH_CHANCE and the catch draw. GetPokeBallWobble draws per shake from WobbleProbabilities and picks the break-free text. Trainer battles block the ball but still use it and the turn. catchRate is exported for pure vectors.
• A catch reloads the Pokémon as LoadEnemyMon does (the wild item roll is redrawn, then the original item restored), keeps HP/status/item/DVs, restores the moves and PP it appeared with, adds it to the party with SetCaughtData, registers it and offers a nickname (InitName default for blank names). NewDexDataText shows when the Pokédex is owned. caughtData is now shared with givepoke.
• catchtutorial runs BATTLETYPE_TUTORIAL: "DUDE" from the ROM as <PLAYER>, no send-out, an automatic catch that keeps nothing, and prompts that continue on their own. Both Route 29 triggers, declining and CatchingTutorialDudeScript's repeat offer run from the ROM.
• Text: TX_LOW ($05) is read as a line break (ItemUsedText needs it).
• UI: PACK lists the Balls pocket (B closes it); the in-battle nickname prompt reuses the naming field. The help dialog and README describe the current milestone.
• Refused: a catch with six Pokémon (the PC is not ported), other balls, non-ball items in battle and a transformed target.
• Stress (local, not committed): 2,160 wild battles, 2,047 catches, no refusals.

Checks: 207 unit tests with the local ROM (82 passed, 125 skipped without it); 54 Chromium/WebKit browser tests; native and reference builds; reference, release and vendor checks; git diff --check. The existing keyboard/D-pad test now steps through the enabled PACK button. No emulator comparison of the capture routines was run. Not committed.

Next: the PC and boxes (catching with a full party, SendMonIntoBox, Bill's call), then player switching and other items in battle, and party and bag menus. The Dude's back picture, ball animation and Pokédex page are not presented.

Overworld foundation recovery: 2026-09-25

Resumed Claude Code session 5d15e2b5-2020-4f49-83e7-eddee08c2315, whose user request was “Complete the overworld foundation.” Its last successful check compared Johto and Ice Path animation with the original VRAM; the session then hit its usage limit. The previous handoff ended at capture and was stale.

Preserved and validated Claude's uncommitted implementation:

• Turn delays, bumps, ice and currents, forced movement, dynamic return warps, elevator menus, pits and arrival timing. Six original-engine traces cover outdoor, connected, ledge, interior, scripted-follow and cave movement.
• Surf, Cut, Strength, Whirlpool/Waterfall interaction paths and cycling rules; Strength boulders can fall into script-controlled holes. Later maps are enabled only inside tests, not in the production map list.
• Working map blocks, changeblock, changemapblocks, rollback and minimap invalidation; cave/water/Kanto encounter tables and wildoff.
• Additional NPC movement, variable sprites and Pokémon icons; all movement command decoding and teleport/fishing/skyfall/shake/flicker presentation.
• Tileset animations on private graphics, grass sprite priority and special map palettes. Johto and Ice Path match 96 original VRAM frames after matching the reference animation state, including the scrolling scratch buffer.

Recovery changes:

• Removed zz-grid.test.ts and zz-explore.spec.ts, temporary diagnostics with no assertions. The grid diagnostic also read a private ROM unconditionally.
• Added grass-priority/transparency and special-palette regressions, and ran all 36 tilesets for 2,048 animation frames each while checking that cached ROM graphics stay unchanged. NUM_TILESETS is verified by check:reference.
• Reconciled PLAN, ARCHITECTURE and NATIVE_PORT with the actual implementation and its remaining gaps.

Validation: 271 unit tests passed with the local ROM. With an absent CRYSTAL_ROM_PATH, 127 passed and 144 skipped. The full Chromium/WebKit run passed 64 tests with 8 intentional WebKit skips; two of those passes were the subsequently removed exploratory test, leaving 62 substantive browser checks. The eight new original-engine comparisons run in Chromium only. Native and reference builds, reference addresses/constants, release isolation and vendor checks passed. No physical-device or CI run was performed. Changes remain uncommitted; no push was made.

The whole foundation milestone is not complete. Continue with:

1. Room decorations: the two decoration specials still implement empty slots; stored choices, ownership and the selection flow are absent.
2. Player bag-menu access to GameWorld.useBicycle(). Its rules are tested, but it is not reachable through the current game controls.
3. Remaining reference acceptance: movement traces exclude the original's map-entry tick and compare position/facing/selected actor positions, not event flags. Most special poses and palettes lack executed-reference or screenshot comparisons. The Celebi forest animation stays on frame one.

Do not clear those checklist items or enable later maps based on the existing movement and rendering tests alone.

Overworld foundation completed: 2026-09-25 (resumed from Codex)

After the recovery above, Codex continued with "Complete the foundation" and stopped during its final validation, before updating this file. It had already marked the milestone done in PLAN.md. Its work:

• Room decorations (world/decorations.ts): the initial bed and Town Map poster, ownership through DecorationAttributes event flags, the bedroom PC's selection and put-away flows (including left/right doll placement), room blocks and variable sprites from saved choices, save validation and migration of saves without decorations. The poster opens the native map.
• Start/Select opens Key Items; the Bicycle follows the native indoor and outdoor rules, with keyboard and touch controls (foundation-browser.spec.ts).
• Movement traces now compare event flags as well as positions. Pose checks compare direction, visibility and vertical offset during scripted movement. An executed LoadMapPals matches 192 native palette cases. Restless Ilex Forest and Tin Tower animation match original VRAM.
• Fixed fishing and flicker pose timing against the original routines.

Claude Code finished the increment. Codex had just added two pose traces. Tree shake passed. Skyfall departure (skyfall_top) failed: StepFunction_SkyfallTop leaves OBJECT_SPRITE_Y_OFFSET at $60 on the tick that ends the command, and the offset persists through later steps. Actors now keep a yOffset. Skyfall, teleports, the fishing bite and jumps clear it, because those routines write the offset again. In the game, only the player uses skyfall_top (Ruins of Alph chambers, Burned Tower), just before a warp. The traces now number fourteen. PLAN and NATIVE_PORT were updated to match.

Checks: 276 unit tests with the local ROM (127 passed, 149 skipped without it); 81 Chromium/WebKit browser tests passed and 19 were skipped intentionally (the engine parity tests run in Chromium only); native and reference builds; reference, vendor and release checks; git diff --check. Not committed.

Next (PLAN section 2): the PC and boxes (a catch with a full party, SendMonIntoBox, Bill's call), then the introduction's naming and clock screens. Also open there: player switching and non-ball items in battle, controlled-seed comparisons beyond damage and type effectiveness, and a continuous opening-quest playthrough with save/reload. Later maps stay gated until their scripts and story are supported.

The PC and boxes: 2026-09-25

The foundation above was committed and pushed as 451464f. This increment continues PLAN section 2 with its first open item.

• party/storage.ts: 14 boxes of BoxMon (a party struct without status, HP and stats), the current box, box names (null is the ROM's "BOX" and number) and 50 PC item stacks. SendGetMonIntoFromBox deposit and withdraw, RestorePPOfDepositedPokemon and CalcLevel are shared helpers. maxPP (ComputeMaxPP) now also serves HealParty.
• scripts/pc.ts: PokemonCenterPC and the house PC. The Pokémon Center PC opens Bill's PC (deposit, withdraw, release, change and name boxes) and the player's PC (item withdraw, deposit and toss, the empty mail box), plus Oak's rating once the Pokédex is owned. The house PC adds DECORATION. Menu strings, lists and texts are read from the ROM (PC table in the profile, verified by check:reference). The generic menu effect gained per-item details (levels, box counts). A notice effect covers unported options without aborting.
• Facing tiles now use the ROM's TileCollisionStdScripts in any direction, as in the original (the old table only fired when facing up).
• Capture with a full party: SendMonIntoBox puts the catch first in the current box, with nickname and BallSentToPCText. A full box keeps the ball (Ball_BoxIsFullMessage). Filling the box sets BATTLERESULT_BOX_FULL. reloadmapafterbattle then queues Bill's call through a new memscript effect (LoadMemScript). It is kept in memory only and runs before scene scripts.
• Save format 3 gained storage and pcItems, migrated by absence.

Not ported: MOVE PKMN W/O MAIL and the stats screen (notices), printing, givepoke with a full party (still refused; unreachable in the open maps), the save messages after a box switch, and the tutorial with a full party and box (the original loops forever; refused).

Checks: 290 unit tests with the local ROM (130 passed, 160 skipped without it); 83 Chromium/WebKit browser tests passed, 19 skipped intentionally (including a new Pokémon Center PC flow in both browsers); native and reference builds; reference, vendor and release checks; git diff --check. Not committed.

Next (PLAN section 2): the introduction's naming and clock screens; player switching and non-ball items in battle; MOVE PKMN W/O MAIL and the stats screen with the party menu; a continuous opening playthrough.

The introduction's naming and clock: 2026-09-25

The PC increment above was committed as 3bdca00. This increment finishes the next open item of PLAN section 2.

• scripts/intro.ts (IntroRun): InitGender, OakSpeech with InitClock and NamePlayer, as a native generator the world runs like a script (GameWorld.beginIntro). Texts, menus, preset names and time strings come from the ROM through a new INTRO profile table verified by check:reference.
• New effects: portrait (Oak's trainer picture or the player's, decoded by PokemonData.trainerPic, playerPic and trainerPalette) and a clock stepper choice.
• Game clock: clock.minuteOffset is added to the device time for time of day, weekday, NPC hours and the daily reset. The host's deviceWeekday became clockDay.
• GameSession.load calls the runtime's optional newGame() when nothing is saved. Saves gained intro (true while pending; a save made then restarts the introduction). The shell's name/character fields are gone.
• Browser fixture specs strip this.world?.beginIntro(); from the served runtime, since their worlds stand in for saved sessions. Fresh-load tests play the intro with tests/intro-driver.ts.

Not presented: palette fades, Wooper's cry, the picture slide while naming, ShrinkPlayer and music. Intentional change: the steppers start at the device time instead of 10:00.

Checks: 298 unit tests with the local ROM; native and reference builds; reference, vendor and release checks; git diff --check; 87 Chromium/WebKit browser tests passed, 19 skipped intentionally (including the new intro flow and a reload during it, in both browsers). Committed and pushed to main as b13fd20.

Next (PLAN section 2): player switching and non-ball items in battle; MOVE PKMN W/O MAIL and the stats screen with the party menu; a continuous opening playthrough.

Remaining opening-milestone items: 2026-09-30

The introduction was committed and pushed as b13fd20 (handoff note eb03a20). Asked to go over what was left and do the quickest items first, this session took PLAN section 2's open items in that order:

1. Evolution decoder: PokemonData.evolutions returns typed level, item, trade, happiness and stat records (Tyrogue's are four bytes). A local-ROM test compares all 251 lists with the pinned evos_attacks.asm. Evolving is not ported.
2. Reference vectors: tests/reference-parity.spec.ts now also calls the original CalcMonStats (300 cases, including levels above 100 for the 999 cap; the trampoline sets BC, DE and HL) and CalcExpAtLevel (all 600 growth/level pairs). A mutation (removing the cap) was confirmed to fail. Symbol loading and the ROM hand-off are shared helpers now. The checklist item stays open for random-driven routines (accuracy, criticals, damage variation, capture).
3. Continuous playthrough: tests/playthrough.test.ts plays a new game from the intro to the catching tutorial through GameWorld inputs with a seeded RNG, reloading a JSON save after every stage. Wild battles are fought, or fled below half HP (fighting everything whited out on Route 29). The rival battle is lost with this seed (CANLOSE), and the story continues. The stages share one world, so run the whole file.
4. Switching and healing items in battle (details in NATIVE_PORT):
   • PKMN/TryPlayerSwitch/PlayerSwitch/BattleMonEntrance with SendOutMonText/WithdrawMonText; AskUseNextPokemon, ForcePlayerMonChoice, DoubleSwitch; Shift mode's OfferSwitch. New party prompt (menu, cancel) answered through battleChoose (a slot or -1), and { action: "pkmn" }.
   • Items pocket: CheckItemContext, ItemEffects dispatch by ROM address, RestoreHPEffect and StatusHealingEffect with the ROM's tables and party menu texts. PARTYMENUTEXT_* values are PRET's ($F0-based; check:reference caught a wrong value).
   • UI: PKMN party list with the SWITCH/STATS/CANCEL submenu (STATS disabled), and ITEM/BALL POCKET tabs in PACK (opens on Items). The keyboard test now passes through the enabled PKMN button.

Known limits: hBattleTurn at the battle menu is not modelled everywhere, so a status heal that would make CalcPlayerStats reapply a penalty to a paralysed or burned enemy is refused. WithdrawMonText treats the multiplicand's leftover high byte as zero. X items, Poké Doll, Full Restore, revives and the stats screen are not ported.

Checks: 324 unit tests with the local ROM (130 passed, 194 skipped without it); 93 Chromium/WebKit browser tests passed, 19 skipped intentionally; native and reference builds; reference, release and vendor checks; git diff --check. Committed and pushed to main as f8e0f9c.

Next (PLAN section 2): controlled-seed comparisons of random-driven battle routines. Then PLAN section 3/4 items: the party and bag menus outside battle (with the stats screen and MOVE PKMN W/O MAIL), evolution itself, and the remaining battle items.

Battle items, hBattleTurn and the stats screen: 2026-09-30

After f8e0f9c/cfd083c, the user asked for the items the previous summary listed as refused or not ported:

• Every non-ball item effect usable in battle, found by reading each item's battle context and ItemEffects entry from the ROM: Full Restore, Revive/Max Revive, Energypowder, Energy Root, Heal Powder, Revival Herb, Bitter Berry, Ether/Max Ether/Elixer/Max Elixer/ MysteryBerry (new move prompt), X items, X Accuracy, Guard Spec, Dire Hit and Poké Doll. ChangeHappiness now reads HappinessChanges (including the wPartyMenuCursor quirk), which also fixed level-ups: LevelUpHappinessMod's larger gain where the Pokémon was caught was missing (new landmark battle option). A second participant set (wBattleParticipantsIncludingFainted) serves revives.
• hBattleTurn at the battle menu: DoBattle ends on the player's turn, full turns end on the enemy's, and chooseTrainerMove now returns enemyTurn when AI_Types, AI_Status or AIDamageCalc ran. The status-heal refusal is gone; the enemy's penalty is reapplied as in the original.
• Stats screen: three pages from ROM strings (STATS profile table, verified by check:reference), TypeNames and GetGender. It opens from BattleMonMenu and from Bill's PC (a stats script effect; world.statsScreen).
• Emulator comparisons: a new reference test runs SendOutMonText (with its final jp BattleTextbox patched to ret in the in-page ROM copy), WithdrawMonText's text routine, RestoreHealth (with wHPBuffer1 set as ItemRestoreHP leaves it), RestorePP, ChangeHappiness and CalcPlayerStats with both turns. A mutation of calcPlayerStats was confirmed to fail. The pure helpers are exported from battle.ts.

Remaining: Kurt's and the Park Ball; the leftover multiplicand byte in WithdrawMonText is assumed zero; no whole battle has been compared with the original.

Checks: 337 unit tests with the local ROM (131 passed, 206 skipped without it); 97 Chromium/WebKit browser tests passed, 19 skipped intentionally; native and reference builds; reference, release and vendor checks; git diff --check. Committed and pushed to main as b13e09e.

Controlled random battle comparisons: 2026-09-30

Resumed from be7575a with a clean working tree. The user requested a bounded increment suitable for a limited session. Continued PLAN section 2 with accuracy, criticals and damage variation; capture stays as the next item.

• tests/reference-random.spec.ts runs the original, unpatched BattleCommand_Critical, BattleCommand_CheckHit and BattleCommand_DamageVariation in the reference emulator. The native commands receive the same ten-byte Link RNG stream. Results include the final RNG stream/cursor, so extra or missing draws are detected too.
• Per browser: 2,560 critical cases (all random bytes at all five reachable stages and held-item exceptions), 2,914 accuracy cases (all stage pairs, Foresight, BrightPowder, bypasses, Dream Eater/substitute and Jump Kick miss damage), and 2,149 variation cases (0/1, rounding, 16-bit boundaries, rejected draws and RNG stream regeneration).
• Shared ROM input and symbol helpers moved from reference-parity.spec.ts into tests/reference-fixture.ts. No ROM bytes or extracted assets added.
• Deliberately changing critical < to <=, accuracy >= to >, and the damage divisor from 255 to 256 each made its comparison fail. All three mutations were reverted; no production gameplay change was needed.

Boundary: these compare isolated commands with controlled random inputs. They do not compare a whole battle, the hardware-timed non-Link RNG, or the Protect text path. Lock-On/Fly/Dig remain refused. This does not enable Link battles or later maps.

Checks: 337 unit tests passed with the local ROM; 131 passed and 206 skipped with an absent ROM. All 103 Chromium/WebKit browser tests passed, with the existing 19 intentional WebKit skips. Native and reference builds, reference, vendor and release checks, and git diff --check passed. No CI or physical-device run. This increment was initially left uncommitted and is included in the subsequent capture-comparison delivery below.

Next: controlled-input capture comparisons (PokeBallEffect and GetPokeBallWobble), then the party/bag menus outside battle, MOVE PKMN W/O MAIL and evolution. Keep section 2's capture checklist open until verified.

Controlled capture comparisons: 2026-09-30

Continued PLAN section 2 from the existing uncommitted accuracy, critical and damage-variation comparisons. Preserved that work and completed its next capture-comparison item.

• tests/reference-capture.spec.ts compares the native capture decision with the calculation inside PokeBallEffect, then its shake loop with GetPokeBallWobble. Per browser: 3,220 capture cases and 7,424 shake cases, including supported ball multipliers, all catch-rate bytes, HP scaling and overflow, status bonuses, equality boundaries, every random byte at each wobble-table entry, every escape count and guaranteed catches. Final rate, caught/escaped outcome, shake count and random-draw count agree.
• Capture uses hardware Random, not BattleRandom. The private in-page ROM copy replaces Random with a supplied-byte reader; the fixture enters after the UI setup and inserts a return after wWildMon is written. Catch arithmetic, multipliers and GetPokeBallWobble remain unmodified.
• Extracted rollCapture and ballWobbles as private production helpers called by pokeBallEffect, without changing gameplay behavior. Added maximum HP 683 beside 342 in the synthetic divide-by-zero regression.
• Deliberately changing success <= to <, formula floor to ceiling, and shake >= to > each failed the comparisons. Mutations were reverted.

Boundary: isolated capture decision/shakes with supplied random bytes, not hardware RNG timing, animation, caught-mon reloading, party/PC writes or a whole battle. No actual item has HELD_CATCH_CHANCE; that effect remains covered by pure vectors. Native division-by-zero refusal remains in place; the emulator fixture omits the hanging HP inputs. No additional balls or maps are enabled.

Checks: 337 unit tests passed with the local ROM; 131 passed and 206 skipped with an absent ROM. All 107 Chromium/WebKit browser tests passed, with the existing 19 intentional WebKit skips. The four new capture tests skip cleanly without a ROM. Native and reference builds, reference, vendor and release checks, and git diff --check passed. No CI or physical-device run. The accuracy, critical, damage-variation and capture comparisons were committed together as 00972ba and pushed to origin/main; the remote commit was verified. This handoff-only follow-up records that delivery.

Next: party and bag menus outside battle, MOVE PKMN W/O MAIL and evolution. PLAN section 2 is complete within its recorded acceptance boundaries.