Add benchmark-integrity audit issue pack with evidence-backed remediation criteria

[Copilot]

This PR adds a strict benchmark audit artifact focused on anti-mimicry evaluation quality. It translates the identified risk areas into ready-to-file GitHub issues with concrete repo evidence and measurable upgrade criteria.

• Scope: benchmark audit issue set

  • Added docs/BENCHMARK_AUDIT_ISSUES.md with 7 issue drafts covering:
    • proxy metric validity
    • benchmark framing accuracy
    • missing ground-truth LoRA/DreamBooth evaluation
    • robustness transform gaps (incl. JPEG/crop)
    • dataset split leakage risk
    • reproducibility metadata gaps
    • weak comparative methodology

• Issue format standardization

  • Each draft includes:
    • Title
    • What is wrong
    • Why it misleads
    • Evidence from repo (file/line references)
    • Proposed benchmark upgrade
    • Acceptance criteria
    • Labels

• Traceability note

  • Added a header note to re-verify line references at filing time to account for line drift across commits.

Example excerpt format used in the new document:

## 4) Robustness suite misses common purification transforms

**Title**
`Robustness testing gap: no JPEG compression or crop transforms in protection-readability benchmark`

**Evidence from the repo**
- Current transform suite: identity, blur, resize(0.75), resize(0.5): `src/auralock/core/style.py:217-231`

**Acceptance criteria**
- [ ] Transform suite includes JPEG and crop variants by name.
- [ ] Protection report contains per-transform metrics for all added transforms.

Original prompt

---

This section details on the original issue you should resolve

<issue_title>audit benchmark</issue_title>
<issue_description>Analyze this repository specifically for benchmark integrity and anti-mimicry evaluation quality.

Create GitHub issues for:

• proxy metrics that may not reflect real anti-mimicry strength
• misleading benchmark framing
• missing ground-truth LoRA/DreamBooth evaluation
• weak robustness testing (resize, JPEG, blur, crop)
• dataset split risks
• missing reproducibility details
• weak comparison methodology

For each issue, include:

• Title
• What is wrong with the benchmark
• Why it can mislead users
• Evidence from the repo
• Proposed benchmark upgrade
• Acceptance criteria
• Labels

Be strict and technical.
</issue_description>

Comments on the Issue (you are @copilot in this section)

• Fixes audit benchmark #5

---

⌨️ Start Copilot coding agent tasks without leaving your editor — available in VS Code, Visual Studio, JetBrains IDEs and Eclipse.