# Neon Intent Candidate Guard

- Run: `run_019_neon_intent_candidate_guard`
- Captured at: `2026-05-25T04:27:29.348Z`
- Target: `https://main.dlwavl00q582x.amplifyapp.com/games/drum-sequencer?song=monkberry-moon-delight-tab&mix=profile&view=trainer&instrument=bass`
- Profile: `lilarcade-neon-ratchet-loop-mix-level-search`
- Result: `passed`
- Review packet status: `needs_followup`
- Deploy: Amplify job `695`, commit `2104997d01fa1f260501bba2c775bed0df2d3f52`, status `SUCCEED`

## Catch

The previous ratchet loop could rank a supported candidate that did not match the requested musical claim. The request said:

`turn the chord part down a little`

But a broad review-order sweep could surface a `bass -0.18` candidate because it scored well under objective guardrails. That is useful exploration data, but it is not support for the chord-down claim.

The guard now infers the claim target as:

- target track: `chord`
- direction: `down`
- source: `intent_text`

After the fix, the live proof only tested chord-down candidates for this default request. No candidate was promoted:

| Candidate | Edit | Failed receipts |
| --- | --- | --- |
| chord -0.05 | `chord 0.16 -> 0.11` | `required_instruments_preserved`, `section_energy_floors_preserved` |
| chord -0.10 | `chord 0.16 -> 0.06` | `required_instruments_preserved`, `section_energy_floors_preserved` |
| chord -0.18 | `chord 0.16 -> 0` | `required_instruments_preserved`, `section_energy_floors_preserved` |
| chord -0.25 | `chord 0.16 -> 0` | `required_instruments_preserved`, `section_energy_floors_preserved` |

The approval-surrogate and durable patch steps were skipped because the packet had no supported review candidate.

## Boundary

This does not prove the current chord level is artistically best. It proves the ratchet will not promote an off-target candidate for a requested claim, and it will not apply a durable mix patch when every matching candidate violates deterministic preservation receipts.