Vision-Language Capability Ledger Data Receipts
Public Draft v0.2 REAL DATA
8 parallel Opus research-agent sweeps yielded ~403 raw papers, deduplicated and hand-arbitrated to 397 unique. Bills 4, 7, 10 ★ NO CLEAN TRIGGER YET (0 clean triggers each). Rebuttal density 22.4%.
Receipts
| Artifact | Link | Purpose |
|---|---|---|
| Bill definitions | bills_draft.md | 12 bills + 6 meta-costs + 3 escape gates + ★ 4, 7, 10 empty-space verification with real fire counts |
| Threat model | purpose.md | Threat model, scope, empty-space hypothesis, cousin-ledger coupling |
| Corpus union JSON | _batch_1_union.json | 397 unique papers (deduplicated from ~403 raw across 8 sweeps), with full metadata |
| Classifier | bill_classifier.py | Regex rule engine + hand-arbitration. Run with --arbitrate-union |
| Aggregator | aggregate_batch_1.py | Deduplicates raw sweep JSONs into the corpus union |
| README | README.md | Reproducibility README with run order |
Real fire counts
| Bill | Cands. | Clean | Rebuttals | Gated |
|---|---|---|---|---|
| 1 — Image-search / web-snapshot contamination | 11 | 9 | 2 | 0 |
| 2 — OCR-extracted-text leakage | 11 | 5 | 4 | 2 |
| 3 — Vision-tokenizer-format brittleness | 31 | 16 | 6 | 9 |
| 4 ★ Causally-faithful vision-grounding mechanism | 32 | 0 | 13 | 19 |
| 5 — Cross-VLM-architecture portability | 21 | 15 | 4 | 2 |
| 6 — Tool-augmented-vision decomposition | 8 | 3 | 2 | 3 |
| 7 ★ Cross-benchmark generalization | 6 | 0 | 3 | 3 |
| 8 — Multi-image / video / interleaved generalization | 15 | 13 | 2 | 0 |
| 9 — Vendor-self-eval independence | 18 | 14 | 1 | 3 |
| 10 ★ Universal vision-task coverage | 5 | 0 | 3 | 2 |
| 11 — Anti-saturation construction | 35 | 27 | 5 | 3 |
| 12 — Distilled-cousin / open-weight VLM audit | 7 | 6 | 1 | 0 |
Public draft v0.2 (2026-05-09). Sweep JSONs live in the source repo at ProjectForty2 public evidence bundle: vision_language/deep_loops/. Target v0.3 lock 2026-Q3.