Gatchan/FF6Portage

Watch

Fork 0

Route checks with ports on: timing-invariant checkpoints, SRAM start points only (fixes #34) #44

Merged

Gatchan merged 4 commits from ports-on-checks into main

2026-11-02 23:52:18 +00:00

Gatchan commented

2026-11-02 22:07:30 +00:00

Owner

Fixes #34 (decision of 2026-11-02, workdocs/decisions.md).

Check definitions: before / after

	Before	After
Semantic checkpoint (`--semantic`, `route_ports_on.*`)	every checkpoint key, including party HP after a battle	timing-invariant facts only: mode, map/position at the goal, party, event bits, scripted-battle outcome (`battles: {...: "won"}`), save slot. Keys in a route's `EXACT_ONLY` (party HP) are compared by exact checks only
Chained routes	`START = "after:<route>"`: frame-exact replay of the previous movie, even with ports on	`after:` removed (run_route refuses anything but `power_on` / `sram:<route>`); chained routes start from SRAM start points made by the interpreter
Mandatory battles	3 attempts (title, Whelk, Marshal, Figaro escape), each route its own loop	`behaviours.win_battle`, 8 attempts (`BATTLE_ATTEMPTS`), retries jittered; attempt 1 unchanged, so interpreter movies keep their inputs
`rom.ports` divergence	reported (frame + first differing byte)	reported and classified against the timing census (E4, E5, E8 sites, US/JP addresses); an unlisted site prints `FINDING`. Still not a failure
`docs/validation.md`	—	validation doctrine: per-call shadow = exact proof; interpreter movies = frame-exact; ports-on (later native) = goal + timing-invariant checkpoints; rom.ports divergences classified. No cycle charging in bridges

No new gate: route_ports_on.* keeps its place; only its definition changed.

New SRAM start point

title_to_mines_save (power-on, the title route, cliffs, mines save point (33, 22) of map $29, save in slot 1; US 23 834 frames, JP 23 368) is the start point of field_menu_tour (now steps off the save point first; 1 602 / 1 304 frames instead of 10 897 / 10 202) and narshe_to_world_map (62 504 / 56 648 frames instead of 80 001 / 80 910). narshe_to_world_map now saves on a fixed world tile (84, 36): the position after "prove control" depended on input timing (JP ports-on ended on (84, 36) vs (84, 35)). Movies re-recorded; title_to_first_battle_win pads and end hashes unchanged (only the checkpoint comment lines changed).

Evidence

Release ctest (all labels, -j2): 61/61 passed, 6 582 s wall (includes regenerating the 4 SRAM start points, 1 988 s of fixtures; cached afterwards).
route_ports_on.* (all ports on = battle/Rand on main), US / JP: title_to_first_battle_win ok/ok, title_to_mines_save ok/ok, field_menu_tour ok/ok, narshe_to_world_map ok/ok, narshe_to_figaro ok/ok.
rom.ports: US first divergence frame 8792, WRAM $131E, PC $C5:06AF -> E8 (sound driver state/code); JP frame 8678, $131E, PC $C5:06B2 -> E8. No finding.
Debug (gcc ASan/UBSan) ctest -LE slow: 24/24; debug rom.ports.*: see comment below. Clang ctest -LE slow: 24/24, no warnings.
The whole debug slow suite is not run: under ASan the emulator is ~12x slower (field_menu_tour movie 187 s vs 15 s), i.e. about 20 h on this machine.

🤖 Generated with Claude Code

Fixes #34 (decision of 2026-11-02, `workdocs/decisions.md`). ## Check definitions: before / after | | Before | After | |---|---|---| | Semantic checkpoint (`--semantic`, `route_ports_on.*`) | every checkpoint key, including party HP after a battle | timing-invariant facts only: mode, map/position at the goal, party, event bits, scripted-battle outcome (`battles: {...: "won"}`), save slot. Keys in a route's `EXACT_ONLY` (party HP) are compared by exact checks only | | Chained routes | `START = "after:<route>"`: frame-exact replay of the previous movie, even with ports on | `after:` removed (run_route refuses anything but `power_on` / `sram:<route>`); chained routes start from SRAM start points made by the interpreter | | Mandatory battles | 3 attempts (title, Whelk, Marshal, Figaro escape), each route its own loop | `behaviours.win_battle`, 8 attempts (`BATTLE_ATTEMPTS`), retries jittered; attempt 1 unchanged, so interpreter movies keep their inputs | | `rom.ports` divergence | reported (frame + first differing byte) | reported and classified against the timing census (E4, E5, E8 sites, US/JP addresses); an unlisted site prints `FINDING`. Still not a failure | | `docs/validation.md` | — | validation doctrine: per-call shadow = exact proof; interpreter movies = frame-exact; ports-on (later native) = goal + timing-invariant checkpoints; rom.ports divergences classified. No cycle charging in bridges | No new gate: `route_ports_on.*` keeps its place; only its definition changed. ## New SRAM start point `title_to_mines_save` (power-on, the title route, cliffs, mines save point `(33, 22)` of map $29, save in slot 1; US 23 834 frames, JP 23 368) is the start point of `field_menu_tour` (now steps off the save point first; 1 602 / 1 304 frames instead of 10 897 / 10 202) and `narshe_to_world_map` (62 504 / 56 648 frames instead of 80 001 / 80 910). `narshe_to_world_map` now saves on a fixed world tile (84, 36): the position after "prove control" depended on input timing (JP ports-on ended on (84, 36) vs (84, 35)). Movies re-recorded; `title_to_first_battle_win` pads and end hashes unchanged (only the checkpoint comment lines changed). ## Evidence - Release `ctest` (all labels, -j2): **61/61 passed**, 6 582 s wall (includes regenerating the 4 SRAM start points, 1 988 s of fixtures; cached afterwards). - `route_ports_on.*` (all ports on = `battle/Rand` on main), US / JP: title_to_first_battle_win ok/ok, title_to_mines_save ok/ok, field_menu_tour ok/ok, narshe_to_world_map ok/ok, narshe_to_figaro ok/ok. - `rom.ports`: US first divergence frame 8792, WRAM `$131E`, PC `$C5:06AF` -> E8 (sound driver state/code); JP frame 8678, `$131E`, PC `$C5:06B2` -> E8. No finding. - Debug (gcc ASan/UBSan) `ctest -LE slow`: 24/24; debug `rom.ports.*`: see comment below. Clang `ctest -LE slow`: 24/24, no warnings. - The whole debug slow suite is not run: under ASan the emulator is ~12x slower (field_menu_tour movie 187 s vs 15 s), i.e. about 20 h on this machine. 🤖 Generated with [Claude Code](https://claude.com/claude-code)

Gatchan added 4 commits

2026-11-02 22:07:33 +00:00

verify: rom.ports classifies its first divergence against the timing census (US/JP: frame 8792/8678, sound driver state $131E, E8) (refs #34 ) fd5e993063

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

verify: routes chain through SRAM start points only; route title_to_mines_save saves at the Narshe mines save point (US 23834 frames SRAM 8C2660AF; field_menu_tour and narshe_to_world_map start from it) (refs #34 ) fe0e5eb365

Timing-invariant checkpoints (battle outcomes; party HP exact only via EXACT_ONLY),
mandatory battles retried up to 8 times with jitter (behaviours.win_battle).
Movies re-recorded: title_to_first_battle_win pads and hashes unchanged.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

docs: validation doctrine (shadow, exact movies, timing-invariant route checks, classified rom.ports divergences) (refs #34 ) 63bdc1f882

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

verify: narshe_to_world_map saves on a fixed world map tile (world_pos was timing-dependent; US 62504 frames, JP 56648) (refs #34 ) 5b4cdaa129

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

Gatchan merged commit b3d3f470b9 into main

2026-11-02 23:52:18 +00:00

Gatchan referenced this pull request from a commit

2026-11-02 23:52:21 +00:00

Merge pull request 'Route checks with ports on: timing-invariant checkpoints, SRAM start points only (fixes #34)' (#44) from ports-on-checks into main

Gatchan referenced this pull request

2026-11-02 23:52:38 +00:00

route_ports_on: semantic checks depend on battle RNG timing and on frame-exact after: replays #34

No reviewers

No labels

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

Gatchan/FF6Portage!44

No description provided.

Rows
Columns