Validation and evidenceEdit on GitHubSource: docs/COVERAGE.md

Line coverage — the measurement of record

The "~96 % line coverage" on the README badge, the README CI (continuous integration) table, the crates.io and PyPI (Python Package Index) front pages, the kshana.dev hero and the technical report is this measurement, rounded to the nearest whole percent. tests/coverage_figure_doc_sync.rs fails if any of those six surfaces states a different figure from the one recorded below, so the next re-measurement moves all of them together or none.

Measured 95.63 % — 37,697 of 39,419 lines
Commit b1d350d (engine 0.27.2 + unreleased)
Run GitHub Actions, ci.yml job coverage, run 35934579709, 2026-09-24
Tool cargo tarpaulin --engine llvm (LLVM source-based instrumentation)
Scope src/, excluding the generated data tables src/*_data.rs and the thin CLI (command-line interface) entry point src/main.rs; tests/ and web/ are not measured
Artifact coverage-lcov (lcov.info, 239 source files) attached to that run
Floor the same job fails below 85 % (--fail-under 85) — the floor is enforced on every push; the figure above is a record, not a gate

The command, verbatim from .github/workflows/ci.yml:

cargo tarpaulin --engine llvm --out Lcov --output-dir coverage \
  --exclude-files 'src/*_data.rs' --exclude-files 'src/main.rs' \
  --exclude-files 'tests/*' --exclude-files 'web/*' \
  --timeout 3600 --fail-under 85

Re-measuring#

Every push to main re-runs the job and attaches a fresh lcov.info. To move the published figure, download the artifact from a green run and sum it:

gh run download <run-id> -n coverage-lcov -D /tmp/cov
awk -F: '/^LF:/{f+=$2} /^LH:/{h+=$2} END{printf "%.2f%% (%d/%d)\n", 100*h/f, h, f}' /tmp/cov/lcov.info

then update the table above and the six surfaces in the same commit; the doc-sync test names any you miss.

History#

The figure has been re-measured, not re-typed: 95.85 % at v0.17.0 (commit 1daa012, 2026-06-14, published as ~96 %), and ~97 % in two earlier release notes (CHANGELOG.md, left as history). paper/kshana-technical-report.md said "near 97 %" until it was brought to this record; it is now the sixth surface the test pins.