# Multimodal Generation Ledger — Bills Draft (v0.1)

> **13 bills + 6 meta-costs + 3 escape gates**, ★ at 5, 8, 11.

| # | Bill | What gets paid | ★ |
|---:|---|---|:---:|
| 1 | Prompt-leakage contamination | Test prompts absent from training-data corpus (LAION-5B, JourneyDB audit). | |
| 2 | Attribute-faithfulness audit | Counting / color binding / spatial layout / negation handling. | |
| 3 | Text-rendering generalization | Held-out text strings (non-trained typography, rare characters). | |
| 4 | Physics-consistency audit | Video objects don't teleport / interpenetrate / violate gravity. | |
| 5 | **★ Causally-faithful generation mechanism** | Intervention experiments show attention causally produces artifact. Predicted empty. | ★ |
| 6 | Cross-resolution / cross-aspect generalization | Trained 1024² generalizes to 4096² / 9:16 / 1:1 / 16:9. | |
| 7 | Strong-baseline classical comparison | GAN / VAE / flow baseline at equivalent compute. | |
| 8 | **★ Cross-modality unified generation** | Same model image + video + audio above clean threshold. Predicted empty. | ★ |
| 9 | Vendor-self-eval independence | T2I-CompBench / GenAI-Bench / VBench independent reproduction. | |
| 10 | Held-out style / prompt distribution | Diverse out-of-distribution prompts beyond training corpus. | |
| 11 | **★ Held-out compositional generalization** | T2I-CompBench / GenAI-Bench / SeedBench-2 held-out splits above threshold. Predicted empty. | ★ |
| 12 | Commercialization-vs-research axis | Closed cloud (Sora, Veo, MJ) vs open-source (SD3, Flux, HunyuanVideo). *B7 bridge test.* | |
| 13 | Safety / NSFW / deepfake / copyright audit | Watermarking, C2PA, model-extraction risks, training-data attribution. | |

## Iteration plan (8 sweeps)
- 1101: Frontier image generation vendor cards (DALL-E 3, MJ v6/v7, SD3, Flux, Firefly, Imagen 3)
- 1102: Frontier video generation (Sora, Veo, Runway Gen-3/4, Kling, Pika, HunyuanVideo, Mochi)
- 1103: Frontier audio generation (Suno, Udio, MusicGen, ElevenLabs v3, Stable Audio)
- 1104: Prompt-leakage / training-data contamination audits (LAION-5B Carlini extraction, JourneyDB audit)
- 1105: Attribute faithfulness + compositional benchmarks (T2I-CompBench, GenAI-Bench, VBench, SeedBench-2)
- 1106: Physics-consistency video audits (VBench-Physics, Sora-violations critique, video-world-model debate)
- 1107: B7 bridge test — closed-cloud vs open-source disclosure axis at generation tier
- 1108: Independent third-party audits + negative-results / hallucination / safety