| """Validate this publication: python -m pip install pyarrow; python validate.py. |
| |
| Checks the preserved source hashes, split counts, Parquet schema, and exact |
| reconstruction of every canonical trace. Uses no network or API credentials. |
| """ |
|
|
| from collections import Counter |
| from hashlib import sha256 |
| import json |
| from pathlib import Path |
|
|
| import pyarrow as pa |
| import pyarrow.parquet as pq |
|
|
| ROOT = Path(__file__).resolve().parent |
| COLUMNS = [ |
| "trace_id", "domain", "instruction", "final_claim_text", "outcome", |
| "failure_kind", "evidence", "trace_json", |
| ] |
| SCHEMA = pa.schema([(column, pa.string()) for column in COLUMNS]) |
|
|
|
|
| def main() -> None: |
| manifest = json.loads((ROOT / "manifest.json").read_text()) |
| provenance = json.loads((ROOT / "provenance.json").read_text()) |
| seen = set() |
| for split, expected_count in [("train", 180), ("test", 120)]: |
| raw_path = ROOT / f"traces.{split}.jsonl" |
| raw_bytes = raw_path.read_bytes() |
| assert sha256(raw_bytes).hexdigest() == manifest["sha256"][raw_path.name] |
| lines = raw_bytes.decode("utf-8").splitlines() |
| traces = [json.loads(line) for line in lines] |
| table = pq.read_table(ROOT / f"{split}.parquet") |
| assert table.schema == SCHEMA, table.schema |
| assert table.num_rows == len(traces) == expected_count |
| assert all(column.null_count == 0 for column in table.columns) |
| rows = table.to_pylist() |
| for line, trace, row in zip(lines, traces, rows): |
| assert row["trace_json"] == line |
| assert json.loads(row["trace_json"]) == trace |
| expected = { |
| "trace_id": trace["trace_id"], |
| "domain": trace["task"]["domain"], |
| "instruction": trace["task"]["instruction"], |
| "final_claim_text": trace["final_claim"]["text"], |
| "outcome": trace["ground_truth"]["outcome"], |
| "failure_kind": trace["ground_truth"]["details"].get("injection", "none"), |
| "evidence": trace["ground_truth"]["details"]["evidence"], |
| "trace_json": line, |
| } |
| assert row == expected |
| assert row["trace_id"] not in seen |
| seen.add(row["trace_id"]) |
| assert dict(Counter(row["outcome"] for row in rows)) == provenance["splits"][split]["outcomes"] |
| assert dict(Counter(row["domain"] for row in rows)) == provenance["splits"][split]["domains"] |
| print(f"{split}: {expected_count} traces; hashes, columns, labels and exact reconstruction OK") |
| for relpath, digest in provenance["preserved_files_sha256"].items(): |
| assert sha256((ROOT / relpath).read_bytes()).hexdigest() == digest, relpath |
| for relpath, digest in provenance["derived_files_sha256"].items(): |
| assert sha256((ROOT / relpath).read_bytes()).hexdigest() == digest, relpath |
| assert len(seen) == 300 |
| print("PASS: 300 unique traces; all recorded file hashes verified") |
|
|
|
|
| if __name__ == "__main__": |
| main() |
|
|
