Trust scoreboard

Legalithm claims to be current, cited and provably maintained. This page is where that gets checked rather than asserted. Every figure below is computed from files in the repository, and every figure that cannot be is marked not yet measured rather than estimated.

4 computed, 5 not yet measured. The blanks are the honest part. An accuracy percentage with no source behind it is worse than no number, so none is published here until it can be sourced.

Measured

EAA pack freshness
4 days since last revision

corpus/eu-eaa/changelog.json, latest entry 2026-08-14

CRA pack freshness
4 days since last revision

corpus/eu-cra/changelog.json, latest entry 2026-08-14

Independent review coverage
0 of 292 subjects reviewed (0%)

lib/eaa/review/provenance.json. A review is a commit to that file.

Corrections published
1 in 5 days of corpus history

Entries in the corpus changelogs whose summary records a correction.

Not yet measured

Shown, not hidden. Each row names what would fill it.

Correction rate, per 100 obligations per quarter
not yet measured

Needs a quarter of history. The corpora were first published in August 2026, and annualising days would invent the number this page exists to refuse.

Ambiguity rate, where the pack declines to determine
not yet measured

No field in either corpus records a declined determination. The hypothesis/claim split exists in the product database and is not read here.

Expert agreement rate
not yet measured

Needs a named external reviewer. Ten were approached on 17 August 2026; the one who replied priced it as paid consulting.

False-positive rate
not yet measured

Needs a labelled set of assessments with known-correct outcomes. None exists.

False-negative rate
not yet measured

Needs the same labelled set.

Every correction Legalithm has published is in the rules changelog. Which frameworks are tracked and what applies when is on the pack roadmap.