Freeze replay checkpoints until frame-stable simulation reaches the requested time. Wait for no-replay simulation before pausing so health and completion assertions do not race the playback clock.

The HCN failure also reproduced with the original production judgement code. Keep judgement rules and assertion tolerances unchanged; validated in the 299-test focused gameplay regression.

Skip offscreen skin updates for taps, long notes and mines while retaining passive judgements, HCN ticks, full long-note spans and rewind recovery. Bound mine visibility on both sides of the lane and avoid zero-duration initial fades.

Plan per-column pool capacities from note residence intervals with a weighted prewarm budget. Keep runtime pool growth available for unusual scroll patterns.

Release builds and 299 focused tests passed. Matched muted captures reduced reverse-scroll tap P99 from 7.4258 ms to 1.8704 ms; dense mine activation and return spikes remain.

test(gameplay): add reusable muted performance diagnostics

Add an opt-in real Player diagnostic for arbitrary chart paths with muted output by default, completed-seek warmup, frame/GC/memory reports and judgement completeness checks. Provide a generic serial case runner with repeats and optional P99 limits.

Document reusable profiling instructions, gameplay design constraints and remaining activation/return spikes. Keep one-off measurement reports, generated captures, host storage and logs out of version control.

Validated with Release builds, 299 filtered gameplay tests, eight muted paired captures, runner threshold/exit-code smoke checks, and documentation/manifest/script checks. The documentation cleanup was checked for stale references and broken links.

Cache the held-head direction delegate and HCN body tick callback to reduce per-update allocations. Read the current hooks from the cached tick callback so pooled drawable rebinding remains correct, with a focused regression test.

Validation: Release builds and 312 filtered regression tests passed. Uncapped IN+HCN comparisons reduced average allocations per update by 32.6%; frame-time results remained variable.

Run real IN conversion with repeatable random lengths and validate both long-note endpoints against the playable beatmap. Cover LN, CN and HCN across three skins with fixed SP and random DP fixtures.

Remove the framework 1000 Hz benchmark cap, keep active and inactive update/draw limits consistent, and report actual pacing, VSync and allocations per update. Extend the generic manifest, script and documentation while retaining muted defaults and documenting the existing backwards-replay scoring limitation and reproduction.

Validation: Release builds, 312 filtered regressions, and 20 subsequent diagnostic configuration/integrity checks passed. Muted desktop captures verified uncapped clocks with VSync off and an explicit 1000/240 Hz capped run. PowerShell syntax, JSON template and whitespace checks passed.

Add invisible-note keysounds and optional yellow outlines with layout and mod support. Keep invisible notes out of scoring and autoplay.

Skip unrelated import data and classify header commands once. Share timing projection without building gameplay scroll segments.

Add focused regression coverage and configurable invisible-note gameplay diagnostics.

Reduce PCM buffer copies, cache per-voice chunk reads, and avoid allocations during sample maintenance. Coalesce cache eviction checks and speed up judgement rewinds.

Add opt-in hot-path benchmarks and regression coverage for audio output, cache lifetimes, and judgement ordering.