Large Goomba blue-coin reward: VERIFIED (2026-10-17, Sol Medium)

Actual gameplay result
- Original course24, mission1, entrance Yoshi, separate intro-complete saves.
- Hold B from polls1600..1649, release, then R1652. Ordinary controller
  input clears the original224-unit large-Goomba collider and ground pounds.
- Large Goomba actor202 ID13 enters death1 with coin payload3 at tick1670.
- It creates blue coin actor290 ID49,param242 at1685 at its own X/Z.
- Walk toward the landed coin at1740..1769. Coin pickup at1754 increases
  the actual course counter from0 to5; both Goomba and coin are removed.
- B2250 still produces a jump; Yoshi returns to normal idle/control in the
  same course24. Both local and installed F: replays complete2450frames,exit0.
- Scout held-jump2600frames also exits0; coin uncollected expires1894 with
  counter0. This is explicitly not a collection pass.
- No gameplay defect was found. Existing native EXE/MAP are unchanged:
  EXE9533D6D00C9D7CB2E5E1663A646635B00B18930AB6FB6844AFCC04757BE3C7A0
  MAP59E3B609815AA01143E9E41DA4082DD679250F8F2F4716C0A60A0F1FE36F3954

What changed
- New replay validator port/tools/check_large_goomba_blue_replay.py links
  the defeated large Goomba to its newly spawned blue reward by time and
  position. Requires actual ground-pound state, exact+5 at coin disappearance,
  nearby player, actor removal, later jump and same-course normal idle/exit0.
- Eight tests reject expired coins, wrong values, unrelated blue coins,
  incorrect enemy rewards, absent ground pound, lost control/course and crash.
- Native Goomba data audit now checks original enemy coin IDs288/289/290/0
  and native coin values1/2/5, alongside existing profile/resource/callbacks.
- Added Blue mode to the existing Watch Reward Verification.ps1 and installed
  Watch Large Goomba Blue Coin Test.cmd on F:. Existing watcher backed up in
  native-checkpoints/native-checkpoint-2026-10-17-blue-coin/previous-launchers.
- Watcher uses separate resettable QA saves, its own runtime INI and the actual
  installed EXE/ROM. The seed matches the successful local fixture SHA256.

Build and tests
- build/build-native-file-menu.cmd passes; native build already up to date.
- CTest52/52 passes,46.68s: build/blue-coin-ctest.log.
- Eight validator unit tests pass: build/blue-coin-verifier-tests.log.
- Native Goomba/egg/coin data audit passes: build/blue-coin-goomba-data.log.
- Local and installed blue-coin validators pass: blue-coin-local-verification.log
  and blue-coin-installed-verification.log. Watcher syntax/parser passes.
- Evidence prefixes: build/port/native-file-menu-baseline-large-blue-held,
  native-file-menu-baseline-large-blue-collect and installed-special-blue.
  PNG screenshots shown before ground pound, blue reward, and after pickup.
- Lossless BMP-to-PNG cleanup retains pixels, CSVs/logs/saves and all source;
  manifest build/blue-coin-image-cleanup.json. No required files removed.

User check
Double-click F:\AI Games\Mario 64 DS\Watch Large Goomba Blue Coin Test.cmd.
It automatically demonstrates the ground pound and blue-coin pickup, then
closes normally. Avoid controller/keyboard input while watching the replay.
This is an automated demonstration with test saves, not a normal play launcher.

Next bounded milestone / model handoff
- Luna Medium is reasonable for remaining routine reward regression checks
  using the established fixtures, replay framework and actor implementation.
  Begin with conditional/current-mission silver-star reward2 release; inspect
  original mission selector before designing inputs. If a new callback/actor
  integration is required, recommend Sol High before that work.
- Five-silver-star completion/Power Star178 integration, broader progression,
  physical Xbox-controller testing and full port completion remain outstanding.
- Overall goal is NOT complete. Standing user instructions permit pausing for
  this recommended switch to Luna Medium. Tracker was already paused on entry.
