audit benchmark

[VoDaiLocz]

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.