# Evidence Index

This page exposes the minimum artifact surface needed to verify benchmark credibility quickly.

## Core Proof Files

- Matrix summary (strict1/2/3, no-locus vs with-locus): [state/matrix/summary_v3.md](state/matrix/summary_v3.md)
- Matrix raw rows (machine-readable): [state/matrix/summary_v3.json](state/matrix/summary_v3.json)
- Chat patch run log: [state/forced_structural_injection_test_chat_patch.md](state/forced_structural_injection_test_chat_patch.md)
- Anchor scoring script: [tools/score_anchor_sensitivity.py](tools/score_anchor_sensitivity.py)
- Prompt protocol: [gemini_structural_attack_prompt_v1.md](gemini_structural_attack_prompt_v1.md)

## Verification Commands

```bash
python tools/score_anchor_sensitivity.py \
  --label chat_patch \
  --no-locus state/matrix/chat_patch_strict3_locus0.json \
  --with-locus state/matrix/chat_patch_strict3_locus9999.json
```

```bash
python tools/score_anchor_sensitivity.py \
  --label harness \
  --no-locus state/matrix/harness_strict3_locus0.json \
  --with-locus state/matrix/harness_strict3_locus9999.json
```

## Reading Order

1. `summary_v3.md`
2. `forced_structural_injection_test_chat_patch.md`
3. `score_anchor_sensitivity.py`
4. `gemini_structural_attack_prompt_v1.md`