# PageRecall tier-0 reproduction log

- Repository: `https://github.com/aadityc91/littraceqa-groundlm2026`
- Checked revision: `2fa753ed4aa2f703b76acd6eec7cbfa3b1a60f80`
- Dataset: `https://huggingface.co/datasets/LitTraceQA/LitTraceQA`
- Command: `python3 reproduce.py --tier 0`
- Run date: September 21, 2026

```text
Reproducing claims up to tier 0
==========================================================
  PASS  T0 dataset integrity
  PASS  T0 dev metrics vs paper Table 1
        all 11 metrics match to 0.0006
  PASS  T0 test submission validates
        Submission is valid: 71 predictions
  PASS  T0 artifact provenance
        14 stamped, 2 pre-provenance legacy artifacts exempted
  skip  T0 shipped package matches HEAD
        dist/ archive absent
==========================================================
4 passed, 0 failed, 1 skipped
Skipped claims need artifacts this checkout does not have; see README.md 'Reproducing'.
```

## Interpretation boundary

Tier 0 recomputes the paper's committed development metrics with the task evaluator, validates the committed test submission and checks provenance records. It does not rerun model calls, PDF parsing or the page-selection experiment. The optional package-archive check skipped because this checkout did not contain the `dist/` archive.