fix(player): detonate mines on a held crossing or a press within PGREAT (LR2)

[nulltask]

Closes #141.

LR2's own changelog (history.txt, the 080114 mine-implementation entry) defines two detonation conditions: passing a mine with the key held ("キー押したまま地雷通過"), or pressing within the PGREAT range ("ピカグレ範囲内でキーを押すと爆発します"). No later entry revises it. The implementation followed losak's secondary writeup instead and used the GOOD window (±40–120 ms depending on rank) for both legs; the primary source wins.

Changes

• Press leg: processLandminePassage now reads the PGREAT window (±8–21 ms) instead of GOOD. goodWindowReachUs is renamed to pgreatWindowReachUs (reads judges[0]; PGREAT sits at index 0 in every ruleset's window set).
• Hold-through leg: anchored to the crossing itself rather than window sampling. PGREAT can be narrower than a frame tick (~16.7 ms), so a held mine could otherwise enter and leave the window between two ticks and be silently retired without detonating. When the window closes behind a mine, it now detonates if the lane was held at the moment of crossing — a kitty key still down counts, as does a non-kitty press within the 120 ms hold-grace window before the crossing. A press landing after the crossing never detonates there: that is the press leg, and it already missed PGREAT.
• Tests: retitled/retuned the negative press test (±18 ms PGREAT at NORMAL rank), added "press inside GOOD but outside PGREAT does not detonate" and a deterministic positive tap test that exercises the crossing leg without depending on tick cadence. The beatoraja SEVENKEYS PGREAT reach assertion ([20_000, 20_000]) replaces the old GOOD one.
• Docs: the mine sections in docs/player-spec.md / .ja.md and docs/bms-spec.md / .ja.md now state the two-condition model and record why losak's GOOD claim is not adopted (primary source conflict, resolved in favor of LR2's changelog).

Notes

• The playlog simulator's mine model (detonate-at-crossing-while-held) was already window-agnostic and needed no change; the residual engine/simulator divergence on the press leg shrinks from ±40–120 ms to ±8–21 ms.

Testing

• vitest run packages/player/src/ruleset/judge.test.ts packages/player/src/index.test.ts — 97 passed.
• The mine test subset ran 5× consecutively with no flakes (guards the tick-cadence concern).
• tsc typecheck and oxlint clean for @be-music/player.

🤖 Generated with Claude Code

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Preview URL	Updated (UTC)
✅ Deployment successful! View logs	be-music-player-demo	f24ed8b	Commit Preview URL Branch Preview URL	Aug 19 2026, 11:07 AM

[github-actions]

Exports Benchmark

• Base SHA: 6c325bc55fa9
• Head SHA: f24ed8b8b5a7
• Comparable cases: 242
• Regression threshold: 8.00%
• Base runs: median of 3
• Head runs: median of 3

Summary

Metric	Value
Improved (>= threshold)	1
Regressed (<= -threshold)	1
Unchanged	240
Median change	+0.23%
Mean change	+0.19%
Head benchmarked cases	242
Head skipped cases	21

Top Regressions

API	Base ops/s	Head ops/s	Change
stringifier.stringifyBms	21398.34	18417.29	-13.93%

Top Improvements

API	Base ops/s	Head ops/s	Change
audio-renderer.writeAudioFile	4806.40	5223.48	+8.68%