Repository metrics

Codecov receives the verified Python host and repository-tool statement/branch report from the main-branch CI run. The native C/Rust and RTL proofs have separate denominators in the coverage guide; the Python badge does not describe hardware acceptance or C++ compiler exception arcs.

The daily PyPI downloads workflow imports the public PyPIStats overall series for loop-timing-witness. It first verifies all seven source workflows at its exact source revision. Its writer stores only downloads/loop-timing-witness.csv on the metrics branch; source code and documentation remain on main. Checkout credentials are disabled, and the write token is passed only to the final Git push.

The CSV contains ISO dates, without_mirrors and with_mirrors counts. It preserves earlier snapshots beyond the provider's rolling history and upserts refreshed dates. Dates with no without_mirrors observation retain an empty field. Missing observations are never converted into zero downloads. A header-only history means the provider has not supplied an observation yet.

404, rate limiting and server failures retain existing history. The scheduled workflow explicitly permits that unavailable state and reports it. Malformed identity, duplicate keys/pairs, invalid dates/counts, broken retained CSV and other HTTP refusals fail without replacing established data. Updates use a flushed, atomic file replacement; unchanged data produces no Git commit.

python tools/pypi_downloads.py --csv downloads/loop-timing-witness.csv
python tools/pypi_downloads.py --allow-missing --project-csv-only

An explicitly downloaded provider response can be imported with --response FILE. An explicitly selected HTTP(S) endpoint can be supplied with --url URL for a relay or transport verification. Such imports validate project identity and values; their source remains the caller's responsibility. The hosted writer uses the fixed official endpoint. Transport tests use actual owned HTTP and certificate-verified TLS sockets; their declared test observations are never published as provider data.

Download badges require actual provider values. A new package may not appear in PyPIStats immediately, and a blocked PePy endpoint does not establish a download count. The sponsor link, package versions and coverage badge describe their own services independently.