# Neon deep exploration proof-window and preset catch

Date: 2026-05-24

Run type: local Riddle Proof runner

Boundary: objective app and audio guardrails only. This does not prove subjective mix taste.

## What failed

A deeper all-current-song Neon exploration sweep found six deterministic findings after the bounded default sweep had already gone green:

- Monkberry Moon Delight (Tab), part 49-64: the declared proof window expected chord activity where that part did not actually have an active chord lane. This was a profile-calibration overclaim, not a taste finding.
- Yakety Yak (Dark), Drop: clipping, peak 1.0138, headroom -0.12 dB.
- Yakety Yak (Dark), Resolve: clipping, peak 1.0136, headroom -0.12 dB.
- Dark Progression, A: clipping, peak 1.0085, headroom -0.07 dB.
- Dark Progression, B: clipping, peak 1.0087, headroom -0.08 dB.
- Midnight Protocol, A: clipping, peak 1.0202, headroom -0.17 dB.

The failing proof sampled 6 available songs, 6 proof-capable songs, 19 parts, and 23 proof windows.

## What changed

The Neon app now validates declared proof windows against active lanes for the sampled part. If a declared window overclaims required active lanes, the proof falls back to generated active-lane windows instead of treating the declaration as universal.

Hot built-in presets were also lowered for Yakety Yak Drop/Resolve, Dark Progression A/B, and Midnight Protocol A.

Finally, LilArcade added a repeatable deep exploration command:

```bash
npm run proof:sequencer:deep-explore -- --url http://127.0.0.1:5184 --output /tmp/neon-deep-explore-command-final-20260524 --max-songs 6 --max-parts 4 --max-windows 2
```

## Final receipt

The final deep command passed with:

- 6 available songs
- 6 proof-capable songs
- 0 skipped songs
- 6 sampled songs
- 19 sampled parts
- 22 sampled windows
- 0 findings
- restoration ok

The useful lesson is operational: keep the default ratchet bounded for speed, but keep a repeatable deeper local sweep for batching deterministic issues before deploy.