Add ImageBench V1 to Section 6 (Benchmark vs. eval / integrity)

[dh7]

Adds one entry to Section 6 (Benchmark vs. eval / integrity).

Why it clears the bar:

• Show your work: 40+ current text-to-image models scored on 192 prompts across 6 categories, with VLM judges + Estimated Preference Score, refreshed as new models ship. Every generated image is published rather than sampled, so failures are inspectable.
• The angle for this section: "no-hidden-outputs" as a concrete integrity property for generative-image evals — the flip side of the cherry-picking that pervades T2I marketing. Complements the paper-heavy entries with a live worked example.
• Open methodology + judge prompts: https://imagebench.ai/methodology-v1.

Happy to move it, reword the one-line note, or drop the 🆕 tag if you prefer.