cert_atlas.submit

Submission records: a portable, checkable claim about a verifier's score.

A leaderboard is only meaningful if entries can be re-derived. A submission therefore carries everything needed to reproduce it — the atlas digest, the exact track, the per-defect outcome — not just a headline number.

Nothing here is signed or authenticated: a submission is a claim, and the leaderboard says so. The atlas digest is what makes it checkable, because anyone can rebuild the corpus and re-score.

SUBMISSION_FORMAT = 'cert-atlas-submission/1'

def track_ceiling(index: dict, track: str) -> float:

The best score attainable in track, computed from the corpus.

In the artifact-only track a defect tagged requires-anchor cannot be caught by anyone: the artifact is internally flawless and the only distinguishing evidence lives outside it. Treating 1.000 as the bar there would fail every honest verifier, so the ceiling is derived from the atlas rather than assumed.

def build_submission( result: dict, *, verifier: str, track: str, url: str = '', notes: str = '', command: Optional[list] = None) -> dict:

Assemble a submission record from a scoring result.

track is artifact-only or anchored; the two are not comparable and the leaderboard keeps them in separate tables.

def validate_submission(sub: dict, *, expected_digest: str = '') -> list:

Return a list of problems; empty means the record is well formed.

A submission at a different atlas digest is not comparable and is rejected rather than silently ranked against entries from another corpus.

def write_submission(path, sub: dict) -> pathlib.Path:

def render_leaderboard( subs: list, *, atlas_digest: str = '', ceilings: Optional[dict] = None) -> str:

Render submissions as Markdown, one table per track.

Ranked by atlas_score, then detection. Entries that fail validation are listed separately as unranked rather than dropped silently.

def load_submissions(directory) -> list: