GameCult/AquaSynth
Watch
9
0
Fork
You've already forked AquaSynth
 0
Code Issues Pull requests 1 Projects Releases Packages Wiki Activity Actions

aqua-feel sync: --shape ramp for slow-swell visuals (measures, never gives a verdict) #34

Merged
aqua merged 3 commits from aqua/sync-ramp-detector into main 2026-11-11 06:24:44 +00:00
Conversation 0 Commits 3 Files changed 5 +152 -12
aqua commented 2026-11-11 04:00:42 +00:00
Member
Copy link

Adds --shape ramp to aqua-feel sync: a detector for slow-swell visual responses (Perlines v2 Vector trails brightens over hundreds of ms; the step detector pairs 0 of 10 hits on it, the ramp detector pairs 9-10).

What it does and does not claim. It finds each swell and prints a point estimate (foot of the line through the 20%/80% crossings) and an interval. Every ramp pair is borderline and a ramp can never fail the gate. Two separate-context Soul passes falsified every stronger claim the first versions made (see SYNC.md "Ramp-shaped visuals"): inside/early (a slow start cannot be bounded from below), late (one outlier raw frame moves the upper bound earlier; wrong at true lags 4-97 ms on heavy-tailed noise, 3% outlier frames, keyframe spikes), and pairing (a bed that swells near a hit takes over). The step path is byte-identical to main on the fixtures and a 126-run synthetic grid.

Tests. 72 checks with ffmpeg (50 without): known-offset synthetic ramps (linear, exponential, smoothstep, quadratic; 30/60 fps) have the true start inside the printed interval where it should; a 120-case sweep never yields a verdict other than borderline/ambiguous/no-visual-onset; noise and drift give no onsets; the outlier-frame and swelling-bed cases; a 0.2 MB real Perlines capture fixture (step finds 0, ramp pairs 3 of 5). Time-based smoothing (not index-based) so variable-frame-rate tracks do not stretch the window; the second Soul pass found no test that pins it, which is recorded, not claimed.

Results on real captures, off-grid hits, and the Playground: https://minds.gamecult.org/aqua/sync-offgrid-ramp-playground/ . Nobody has watched or listened to any of it.

🤖 Generated with Claude Code

Adds `--shape ramp` to `aqua-feel sync`: a detector for slow-swell visual responses (Perlines v2 Vector trails brightens over hundreds of ms; the step detector pairs 0 of 10 hits on it, the ramp detector pairs 9-10). **What it does and does not claim.** It finds each swell and prints a point estimate (foot of the line through the 20%/80% crossings) and an interval. **Every ramp pair is `borderline` and a ramp can never fail the gate.** Two separate-context Soul passes falsified every stronger claim the first versions made (see SYNC.md "Ramp-shaped visuals"): inside/early (a slow start cannot be bounded from below), late (one outlier raw frame moves the upper bound earlier; wrong at true lags 4-97 ms on heavy-tailed noise, 3% outlier frames, keyframe spikes), and pairing (a bed that swells near a hit takes over). The step path is byte-identical to main on the fixtures and a 126-run synthetic grid. **Tests.** 72 checks with ffmpeg (50 without): known-offset synthetic ramps (linear, exponential, smoothstep, quadratic; 30/60 fps) have the true start inside the printed interval where it should; a 120-case sweep never yields a verdict other than borderline/ambiguous/no-visual-onset; noise and drift give no onsets; the outlier-frame and swelling-bed cases; a 0.2 MB real Perlines capture fixture (step finds 0, ramp pairs 3 of 5). Time-based smoothing (not index-based) so variable-frame-rate tracks do not stretch the window; the second Soul pass found no test that pins it, which is recorded, not claimed. Results on real captures, off-grid hits, and the Playground: https://minds.gamecult.org/aqua/sync-offgrid-ramp-playground/ . Nobody has watched or listened to any of it. 🤖 Generated with [Claude Code](https://claude.com/claude-code) 
Perlines v2's Vector trails answers a sound with a brightening over hundreds of ms;
the step detector found no visual onset in any of its takes. The ramp detector takes
the rise over 300 ms as novelty, finds each rise's quiet level and 20%/80% crossings,
and reports the foot of that line with the interval it can defend: from before the
quiet minimum (backed off by the valley-to-20% time, which a property test over three
ramp shapes and decaying tails showed necessary) to the 20% crossing. Tests: synthetic
ramps at known offsets (linear, exponential, smoothstep; 30 and 60 fps), noise and drift
give nothing, and a 0.2 MB real Perlines capture (step finds 0, ramp pairs 3 of 5).

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
A Soul pass falsified the first ramp detector: the +-60 ms centred average crossed
before the picture did (fast rises called detectably late inside the window), the
smoothing ran over samples (stretched on variable-frame-rate tracks), and the lower
bound was a heuristic presented as a bound (slow-start shapes called early or inside
while in sync). Now: time-based smoothing, hiS from the raw frames, and a ramp can
be audio-late-detectable or borderline only. Adds the Soul's fast-rise, VFR and a
120-case sweep over shapes, rises, decays, rates, noise and offsets as tests.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
A second Soul pass falsified the late claim: one outlier raw frame before the true start
moves the upper bound earlier (wrong at true lags 4-97 ms on heavy-tailed noise, 3% outlier
frames, keyframe-style spikes), and a bed that swells near a hit takes over the pairing.
After the first pass removed the early and inside claims, nothing is left that a ramp can
soundly assert about sync, so every ramp pair is borderline and never fails the gate. The
estimate and interval are printed for a person to read. Tests: the outlier frame and the
swelling bed, both with a true lag the old verdict got wrong.

Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
aqua merged commit 9f24e0b42b into main 2026-11-11 06:24:44 +00:00
Sign in to join this conversation.
Reviewers
No reviewers
Labels
Clear labels
No items
No labels
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
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
GameCult/AquaSynth!34
No description provided.