# Correction exchange — ready for an author, not yet a user-study result

The participant's unprepared correction is pending. No preference, artistic judgment, time saving or comparison result has been supplied by the participant. Automated checks are engineering evidence only.

## The exchange

1. The participant describes what the current isolation interpretation gets wrong, in their own words.
2. The assistant proposes a specific executable condition. Preserve the participant's words separately. Identify every added assumption, selected mesh, surface category, clearance, scope and exception.
3. The participant corrects or accepts that operational meaning. If it is outside the supported vocabulary, record the gap and implement a new measurement or leave it unresolved. Do not substitute a nearby meaning without disclosure.
4. Keep a substantial native camera revision; request another revision carrying the same condition.
5. The participant supplies an intentional exception or changes their mind. Preserve the earlier rule in history; apply the exception only to the specified interval. A measured violation in an exempt frame is still a violation, deliberately permitted.
6. Review the moving results with the participant. Record whether the intended relationship survived, whether the exception had the intended effect, and whether controlling it felt worthwhile.

## Supported operational vocabulary

Subject: head/hair convex envelope; union of projected triangles for head, hair, suit and backpack; backpack projected triangles. The figure option binds those four named meshes, not every facial accessory. Surroundings: nearest included rock and steep-terrain surfaces, or rock objects with terrain still participating in occlusion. Flat terrain remains allowed. Clearance: 0–4% of width at a 320×180 raster; zero requires no overlapping pixel centers. Scope: inclusive local frames 1–24. Optional inclusive exception. Priority: required or preferred. Preferred conditions rank fewer violating active frames first but do not certify that the condition passed.

Measurements cover every frame of every projection-feasible camera in the fixed 216-camera pool. This enables changing temporal scope without treating formerly unmeasured frames as passing. Camera search outside that family, new objects, emotional meaning, shadows, vegetation and contact semantics are not implemented by this vocabulary.

## Fair comparison

A subsequent comparative session must use a competently configured conventional native workflow with persistent constraints, the same source, equivalent scene access and the same operational condition. It must not simulate a competitor that forgets decisions. Record setup, repeated explanations/corrections, repair work and participant assessments. Counterbalance workflow order and retain failures. Report assistant intervention and preconfigured assets. A pilot with one participant cannot establish general superiority.

The existing direct-constraint calculation shares the predicate, pool and ranking. Matching cameras checks consistency; it is not an independent artist benchmark. The conventional workflow session has not been run.

## What is stored

The project JSON retains literal notes, explicit condition values, source and measurement hashes, revision ancestry and the selected camera. The per-frame display distinguishes enforced passes, violations, exceptions and out-of-scope frames. Native render jobs record the exact evaluated condition. The participant may leave an outcome unresolved. No inferred satisfaction score is generated.

## Measurement bounds

There are 186 projection-feasible paths, with 24 frames each: 4,464 frame observations and six subject/surroundings combinations per observation. Distances at 0.040625 are lower bounds from a 13-pixel search radius; the control supports at most 4%. These raster estimates are not subpixel or perceptual guarantees. The legacy head/hair rule reproduces the prior eligible sets exactly (4 approach, 6 crossing).

The relation rays are cached. A new render verifies the source hash, the condition evidence hash, the four character meshes against their measured source positions, and the existing geometry/contact/visibility checks. It does not recast all background rays or certify every environmental deformation.

## Current engineering verification

[Verification record](verification.json) and [fresh native render receipt](../direction/direction-7d90492fa1b122225397f36c/receipt.json). The new film contains 24 original-performance frames, freshly rendered in approximately 162 seconds. All native checks passed; maximum deviation across the four character meshes was 0.000040 m. The film used an agent-selected figure/rocks condition with frames 10–14 exempt; those exempt frames also passed, so the film does not demonstrate a necessary exception. A separate source-measurement and interface check verified that local frame 22 on approach camera y20-d3-e3-s-0.12 remains a failure when intentionally exempted. The software suite passes 27 tests. None of these choices is participant data.