# Vision-Language Capability Ledger Ledger — Reproducibility data (v0.2)

**Real-data ledger** produced via the full deep-loop methodology
(8 parallel Opus research-agent sweeps + hand-arbitration).

## Files

- `bills_draft.md` — v0.2 bills definition (12 bills, 6 meta-costs,
  3 escape gates, ★ at Bills 4, 7, 10)
- `purpose.md` — threat model, scope, empty-space hypothesis
- `_batch_1_union.json` — deduplicated union of 8 sweeps;
  397 unique papers
- `bill_classifier.py` — regex + hand-arbitration logic
- `aggregate_batch_1.py` — sweep aggregator
- `README.md` — this file

## Real-data summary

```
8 sweeps × ~49 papers/sweep avg = ~403 raw → 397 unique
Bills 4, 7, 10 ★ NO CLEAN TRIGGER YET:
  Bill 4  — 32 cands / 0 clean / 13 rebuttals / 19 gated
  Bill 7  — 6 cands / 0 clean / 3 rebuttals / 3 gated
  Bill 10  — 5 cands / 0 clean / 3 rebuttals / 2 gated
Rebuttal density: 22.4% (89 papers)
```

## How to reproduce

```bash
python3 aggregate_batch_1.py
python3 bill_classifier.py --arbitrate-union
python3 bill_classifier.py --benchmark
```

## Status

**v0.2 RELEASED 2026-05-09**. Target v0.3 lock 2026-Q3.

## Source ledger directory

`ProjectForty2 public evidence bundle: vision_language/`