Datasets:
File size: 4,859 Bytes
2f49b29 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 | #!/usr/bin/env python3
"""Report whether the current POSA Bench bundle is preview- or rank-ready."""
from __future__ import annotations
import argparse
import hashlib
import json
from pathlib import Path
BENCHMARK = Path(__file__).resolve().parents[1]
REVIEWED = BENCHMARK / "review" / "results"
PUBLIC_MANIFEST = BENCHMARK / "huggingface" / "dataset" / "preview-v0.1.0-manifest.json"
FROZEN_RELEASE = BENCHMARK / "releases" / "preview-v0.1.0"
EXPECTED_MODELS = 4
REQUIRED_SEEDS = {17, 29, 41}
def digest(path: Path) -> str:
value = hashlib.sha256()
with path.open("rb") as handle:
for block in iter(lambda: handle.read(1024 * 1024), b""):
value.update(block)
return value.hexdigest()
def verify_manifest(manifest_path: Path, root: Path, source_field: str = "path") -> tuple[bool, str]:
if not manifest_path.is_file():
return False, f"missing {manifest_path.relative_to(BENCHMARK)}"
manifest = json.loads(manifest_path.read_text(encoding="utf-8"))
errors = []
for entry in manifest.get("files", []):
source = root / entry[source_field]
if not source.is_file():
errors.append(f"missing {entry[source_field]}")
elif source.stat().st_size != entry["bytes"] or digest(source) != entry["sha256"]:
errors.append(f"hash mismatch {entry[source_field]}")
return not errors, "; ".join(errors) if errors else f"{len(manifest.get('files', []))} hashes verified"
def check() -> tuple[list[tuple[str, bool, str]], list[tuple[str, bool, str]]]:
reviewed_paths = sorted(REVIEWED.glob("*.reviewed.json"))
reviewed = [json.loads(path.read_text(encoding="utf-8")) for path in reviewed_paths]
frozen_ok, frozen_note = verify_manifest(FROZEN_RELEASE / "manifest.json", FROZEN_RELEASE / "snapshot")
public_ok, public_note = verify_manifest(PUBLIC_MANIFEST, BENCHMARK, "source")
tracks = {item.get("track") for item in reviewed}
statuses = {item.get("eligibility", {}).get("status") for item in reviewed}
required_quality = {
"semantic_claim_recall", "critical_fact_coverage_global", "safety_rate", "strict_pass_rate"
}
preview = [
("Frozen baseline", frozen_ok, frozen_note),
("Curated public manifest", public_ok, public_note),
("Four reviewed candidates", len(reviewed) == EXPECTED_MODELS, f"found {len(reviewed)}"),
(
"Fixed-context isolation",
tracks == {"fixed-context-common-runtime"},
", ".join(str(value) for value in sorted(tracks)),
),
("Explicitly unranked", statuses == {"unranked-preview"}, ", ".join(str(value) for value in statuses)),
(
"Reviewed metric completeness",
all(required_quality <= set(item.get("quality", {})) for item in reviewed),
"four publication metrics required",
),
]
seeds_by_run = {
item.get("run_id"): set(item.get("protocol", {}).get("seeds", [])) for item in reviewed
}
repetitions_ok = bool(reviewed) and all(
int(item.get("protocol", {}).get("repetitions", 0)) >= 3
and seeds_by_run[item.get("run_id")] >= REQUIRED_SEEDS
for item in reviewed
)
human_review_ok = bool(reviewed) and all(item.get("review", {}).get("official") is True for item in reviewed)
thermal_ok = bool(reviewed) and all(
item.get("protocol", {}).get("thermal_reset_verified") is True for item in reviewed
)
continuous_pss_ok = bool(reviewed) and all(
item.get("protocol", {}).get("continuous_pss_sampling") is True for item in reviewed
)
official = [
(
"Three seeds per model",
repetitions_ok,
"; ".join(f"{run}: {sorted(seeds)}" for run, seeds in seeds_by_run.items()),
),
("Two blinded humans plus adjudication", human_review_ok, "current artifacts are single-AI review"),
("Verified thermal resets", thermal_ok, "not recorded in current protocol"),
("Continuous peak-PSS sampling", continuous_pss_ok, "current values are post-run observations"),
]
return preview, official
def main() -> int:
parser = argparse.ArgumentParser()
parser.add_argument("--target", choices=("preview", "official"), default="preview")
args = parser.parse_args()
preview, official = check()
sections = [("Preview publication", preview)]
if args.target == "official":
sections.append(("Official ranking", official))
for title, gates in sections:
print(title)
for name, passed, note in gates:
print(f" {'PASS' if passed else 'BLOCKED':7} {name}: {note}")
selected = preview if args.target == "preview" else preview + official
return 0 if all(passed for _, passed, _ in selected) else 1
if __name__ == "__main__":
raise SystemExit(main())
|