§ SAMPLES · EVAL CORPUS

The filings AYVES is tested against.

Every entry links to the entity's own source page. Ground-truth rows carry a hand-labelled expected.json used by CI evals every time a prompt or model string changes. Filings without ground truth are used for silent-corpus regression only.

Ground-truth fixtures · 1 filings

Rio Tinto plc
FY2024 · AU_CBCR · 5.8MB
extractive · F6

Silent corpus · 6 filings

Woolworths Group Limited
FY2023 · voluntary · 0.4MB
FMCG · retail · F9
Shell plc
FY2023 · voluntary · 4.9MB
energy · F13
TotalEnergies SE
FY2022 · EU-CBCR · 1.5MB
energy · F15
Equinor ASA
FY2023 · voluntary · 2.0MB
energy · F16
Anglo American plc
FY2023 · voluntary · 8.0MB
extractive · F17
Glencore plc
FY2023 · EU-CBCR · 1.5MB
extractive · F18

The six ship-block metrics (jurisdiction recall, spurious rows, financial-error rate, employee-error rate, Zod first-pass rate, format-detection rate) are computed nightly against the ground-truth set.