# QA3 ABI-aligned coverage-oracle evidence

This directory is generated by `benchmark/run_qa3_coverage_campaign.py`.
It contains the selected and per-round edge sets needed to recompute
the summary without trusting terminal output.

## Protocol

- Rounds: 3
- Inter-round delay: 1000 ms
- Stability policy: `intersection`
- Input mode: `stdin`
- Terminal status policy: `normal-only`
- Campaign observations: 60
- Compatibility-probe observations: 2
- Oracle mode: `one-shot` (1 fallback)

## Observed result

- Baseline intersection/union: 934/935 edges.
- Strict stability passed: `False`.
- Unstable inputs/events: 20/57.
- `nominal`: 19 outputs, 19 normal, 0 terminal-excluded, 19 landing hits, 1374 seed-novel edges.

The bundled corpus has only the `nominal` label. These numbers validate
the coverage oracle and do not establish a solver-strategy speedup.

## Files

- `raw-campaign.json`: round order, every per-input replica edge set,
  and the selected intersection.
- `summary.json`: derived stability, landing, and complementarity metrics.
- `SHA256SUMS.txt`: content manifest for all evidence files.

## Reproduction parameters

- AFL binary: `benchmark/public/bin/google-fts-afl/xml_read_fuzzer`
- Corpus: `benchmark/public/seeds/google-fts/xml_read_fuzzer`
- Seed: `benchmark/public/seeds/google-fts/xml_read_fuzzer/seed_01.xml`
- afl-showmap: `afl-showmap`