alex/flybrain
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

docs: the row 61 audit, a trainer's challenge text walled the forest's corridor

Browse source
This commit is contained in:
acamilo 2026-09-23 22:45:08 +00:00
parent 0117a85354
commit d6597b3e98
1 changed files with 78 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

78
infra/docs/macros-traps.md
View file

@ -2967,3 +2967,81 @@ survey and the three ROM tests, which is a deviation from the ethos check's lett
- `flysim --print-compatibility`: byte-identical to the base on both bases this branch has had:
648 bytes, `4929f340...9ebd9` on v0.5.5 (`7784a9d`), and 648 bytes, `8ce67b97...a8f68` on
`4d82f7d` (v7). Decoder, reward catalog, adapter version and roles untouched.
## 2026-09-23, row 61: the forest's south gate, in and out
### What was live
Map 50 (`VIRIDIAN_FOREST_SOUTH_GATE`, `$32`) and Route 2's south half, rung 9, v0.5.5, about 20:45
to 21:05 UTC: `GO OBJECTIVE` into the gate (0.3 s), `GO OUT` back onto Route 2 (0.2 s), `GO WARP`
back out of the forest, `GO OBJECTIVE blocked` in the forest, `GO ROUTE` refused now and then; no
reward and about eight new tiles in twenty minutes. The checkpoint is the fly at (5, 1) in the
gate, under the forest's doorway, objective Pewter City; `next_hop` answers the forest, correctly.
### The survey: what walled the road
The route survey from the checkpoint (`FLY_PROBE_CATCH=route`, uniform choice per hold,
`FLY_PROBE_RNG=7`, 72,000 frames) walks the live ring and never leaves rung 9. Caught at the
refusal (`FLY_PROBE_CATCH_FRAME`, `FLY_PROBE_WHOLE=1`): the forest's only road to the north gate
is the corridor at x = 1-2 from row 22 up to row 0; the Bug Catcher of
`EVENT_BEAT_VIRIDIAN_FOREST_TRAINER_2` stands on (2, 18) facing west, and **(1, 18) was in the
pushed ledger**. It was written at frame 16,389: a `GO ITEM` walk stepped onto (1, 18), the trainer
took the joypad, and his text closed onto five frames with no box, no script bit and
`wCurOpponent` zero (`StartTrainerBattle` runs after `DisplayTextID`'s close-down). Row 58's held
entry was decided on the first of them. The pushed ledger has no window, so the north gate had no
road for the session; `GO OBJECTIVE` walked to the nearest reachable tile, (6, 1), a dead end, and
was blocked, and the last tiers walked the fly back to the south gate and Route 2. Every base
survey arm walls (1, 18); only a fly that has to come back up the corridor is trapped by it.
| # | trap | trigger | test | fix, or why it is left |
| --- | --- | --- | --- | --- |
| 61 | the frames between a sighted trainer's challenge text and `StartTrainerBattle` read as the fly's overworld: a pad is dealt, ground recorded, and row 58's held push-back written, walling the tile the trainer fired on for the session | any trainer whose line of sight a walk crosses; Viridian Forest's (1, 18), the one free tile of the corridor to the north gate | `a_trainers_challenge_is_the_cartridges_until_its_battle_is_over`, `the_frames_between_a_trainers_text_and_its_battle_deal_no_pad_and_record_no_ground`, `a_trainers_challenge_does_not_wall_the_road_to_the_forests_north_gate` (ROM) | **fixed** in the macro seam: `state::trainer_engaged` reads `wStatusFlags7` bit 3 (`BIT_TRAINER_BATTLE`, set by `CheckFightingMapTrainers`, cleared at `.battleOccurred`); `PokeState`'s `scene` is `Unknown` on an overworld frame with it set and `scripted` is true. `controllable` and `scene::detect` unchanged. `docs/design/macros.md` 12.24 |
| 61b | the south gate deals `GO OUT` (back to Route 2) beside the forest door the objective's road takes | the "a room has to be leavable" tier in a gate whose way on is a passage | -- | **left**: a way back is the fly's choice; with the corridor open no survey stays on it |
### Before and after
The route survey, 72,000 frames (20.1 brain minutes), base `main` vs branch:
| driver | base | branch |
| --- | --- | --- |
| uniform, seed 7 (the live ring) | rung 9, 1,202 tiles, north gate never, (1, 18) walled | **BOULDER BADGE**, 1,647 tiles, north gate f20273, Pewter f23756 |
| uniform, default seed | rung 10, 1,356 | rung 11, 1,476 |
| `GO OBJECTIVE` preferred | rung 11, 1,529 | rung 12, 1,492 |
| `GO OUT`, `GO OBJECTIVE` preferred | rung 11, 1,376 | rung 11, 1,446 |
| uniform, seed 12345 | rung 11, 1,765 | identical |
(1, 18) is walled on every base arm and on no branch arm. The ROM-gated run (the seed-7 driver,
40,000 frames): base walled at frame 16,389, rung 9, **fails**; branch: north gate f20272, Route 2
f21209, Pewter f23755, passes.
The stub-readout trap hunt (30 brain minutes, service frame since FND-01; `main` `4d82f7d` vs
branch): both arms rung 10, 362 distinct tiles, 147 macros all done, 11 of 113 windows flagged.
**The stub does not walk the ring**: it passes the corridor northward and never comes back, so
the only difference is 134 overworld frames the branch reads `Unknown`, the trainer gaps. The
survey and the ROM test are the reproduction.
### Other gates
The north gate's `GO OUT` is toward Pewter (tier 2) and its forest door is withheld; the Route 2
gate, Diglett's Cave's Route 2 house and the Route 22 gate are off the graph (every door `LAST_MAP`
but Diglett's passage). The Route 22 gate is the one building with `LAST_MAP` on both sides of two
different maps (its script sets `wLastMap` by row: under 4 is Route 23); `outdoor_of` cannot name
both and will need a per-door answer when a rung routes through it. The trainer gap is not a gate
fact: every sighted trainer has it.
### Overlap
Row 59 (`fix/loop-row59` `cf10ffd`, unmerged) found the same five frames on Route 3 and holds the
push until thirty frames of overworld. From this checkpoint its commit alone also passes the ROM
test; the two compose (flybrain-gb 429/0 and the ROM test with both). This row reads the fact at
the seam, so the gap also deals no pad and records no ground.
### Gates
- `cargo test --release -p flybrain-gb` with `FLY_ROM`: 428 + 27 passed, 0 failed.
- `cargo test --release -p flysim --no-fail-fast` with `FLY_ROM`, `FLY_DATASET` and the row-61
checkpoint: 185 passed, 0 failed (the integration service test passed this run).
- `cargo clippy --workspace --all-targets -- -D warnings`: clean.
- `npm test` 663 passed; `npm run typecheck` clean; `infra/tests/lint.sh` ALL CHECKS PASSED.
- `flysim --print-compatibility`, raw and macros: 648 bytes, sha256 `8ce67b97...a8f68`, the same
as `main`. Decoder, reward catalog, adapter version and roles untouched.