# Blind Human Evaluation Form — Pilot v0.1

**Do not reveal model/provider identity before this form is locked.**  
Use headphones in a quiet environment whenever possible. Play the full clip at least twice, but do not frame-step until after the first-pass impression is recorded.

## Identification

- Benchmark version: `pilot-v0.1`
- Test ID:
- Blind clip ID:
- Rater ID:
- Headphones used: YES / NO

## Part A — Objective observations

Record what is observable, not what you think the model intended.

- Required event present? YES / NO / UNCLEAR
- If applicable, event count: 0 / 1 / 2 / 3+
- Correct designated actor? YES / NO / UNCLEAR
- Source perceptually clear? YES / NO / UNCLEAR
- Required audible/silent condition correct? YES / NO / UNCLEAR
- Required bystander reaction present? YES / NO / N/A / UNCLEAR
- Reaction directed toward correct target? YES / NO / N/A / UNCLEAR
- Visible gas/smoke/cloud/particles? YES / NO
- Defecation depicted or clearly implied? YES / NO

### FB-C02-specific hard checks

- Exactly one fart event? YES / NO / UNCLEAR
- Event occurs approximately in the 2.5–4.0 s window? YES / NO / UNCLEAR
- Event is attributable to woman in red jacket at rear-left? YES / NO / UNCLEAR
- Other three adults gradually direct attention toward her afterward? YES / NO / UNCLEAR
- No dialogue? YES / NO
- No music? YES / NO

## Part B — Perceptual scores

Use `0, 1, 2, 3, 4` or `N/A` only if the canonical spec marks it non-applicable.

| Code | Dimension | Score |
|---|---|---:|
| A | Acoustic Condition Fidelity | |
| T | Event / AV Temporal Alignment | |
| L | Source Localization | |
| P | Physical / Body-Motion Plausibility | |
| I | Instruction Adherence | |
| S | Social-Causal Response | |
| C | Temporal / Identity Coherence | |

## Part C — Failure flags

Mark all observed codes:

`F01 F02 F03 F04 F05 F06 F07 F08 F09 F10 F11 F12 F13 F14 F15`

Refer to `protocol/FAILURE_TAXONOMY.md` for definitions.

## Part D — Confidence and note

- Confidence: 1 low / 2 medium / 3 high
- Short note (optional):

## Lock

- Locked timestamp UTC:
- After locking: **do not revise because of model identity.**