huggingface/test-results/README.md

metadata

license: other
license_name: proprietary
license_link: https://github.com/szl-holdings/a11oy/blob/main/LICENSE
tags:
  - benchmark
  - receipts
  - governance
  - mathcomp
  - mirror-not-canonical
pretty_name: A11oy staged test-results schema

A11oy test-results — staged dataset schema

This directory defines the future SZLHOLDINGS/a11oy-test-results dataset layout. It is a schema and manifest only in this revision.

GitHub remains canonical. Hugging Face is a generated mirror for review.

Current claim status

• No live benchmark score is claimed.
• No leaderboard metric is claimed.
• No benchmark corpus is redistributed here.
• No model-index metrics are published.

Competition-math benchmark scoring remains staged until corpus digest, receipts, reproducible tooling, and judge agreement are present.

Future dataset layout

README.md
MANIFEST.json
benchmark-map.json
schemas/manifest.schema.json
schemas/result-row.schema.json
schemas/receipt-envelope.schema.json
samples/staged/*.jsonl
results/*.jsonl
receipts/*.jsonl

Only schema examples or receipt-backed staged dry-run artifacts may appear before a sealed run exists. Real results require:

1. immutable corpus digest;
2. raw-score reporting;
3. three-judge panel;
4. append-only receipt chain;
5. unsupported-claim rejection;
6. GitHub CI validation.

Validate the current staged manifest with:

npm run hf:test-results:audit
npm run benchmark:audit