audit
The evidence behind the claims.
This section is not written by hand. scripts/extract-lw.mjs reads the repository's own files, recomputes their figures, cross-checks them against each other, and writes the data every page here renders. A number that stops agreeing with its source stops the build.
The census
118 PHP constructs, each with one verdict and the batch it was proven in. The grid is the whole claim in one screen.
Docs coverage
All 105 Laravel documentation pages with the verdict the project gives them, plus the CPU measurements behind the budget.
Refusals
The 64 constructs that fail loud, where they cluster by batch, and the 5 documentation pages with no compiled path.
The ledger
The shape of the project's own audit ledger — 20 suggestions, the severity spread, and how many are closed — without republishing its item text.
Method
15 build-time checks, the exact files and hashes read, and an explicit list of what this audit does not claim.
Every figure
The whole extraction on one page: every scalar in the payload, every row, every check — for reading and for printing.
66b65a03f (v0.4.10), extracted 2026-11-08. The documentation and the landing page describe ba1e9a5 (v0.4.10); the audit kept the commit it was taken from, so its figures stay reproducible.What this audit is not
It is not a review of the compiler's correctness — the parity harness and the test suite own that. It is a review of the repository's published claims: whether the numbers in the README, the map and the coverage index still describe the code, and whether the wording around them says more than the evidence does.