Per-tag score trends
GENERATED from evals/scores-by-version.json
by scripts/eval/build-score-trends.py — do not hand-edit; regenerate after each ledger row.
Numbers are per-tag scores as recorded per release (evalevalRunning the model against a held-out golden dataset and computing per-component F1, exact-match, calibration, and resolved-coordinate error. sets, channels, and quantization evolve across eras — adjacent columns are comparable, distant ones directional; the dated ship-gate docs carry each column's exact conditions). "—" = not measured that release.
US
| tag | 0.1.0 | 0.2.0 | 3.0.0 | 4.2.0 | 4.3.0 | 4.4.0 |
|---|---|---|---|---|---|---|
| micro | — | — | — | 84.8 | 85.1 | 86.1 |
| streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels. | — | — | 26.6 | 76.2 | 75.5 | 77.9 |
| street_prefix | — | — | — | 64.9 | 93.6 | 93.6 |
| street_suffix | — | — | — | 48.8 | 96.6 | 96.6 |
| house_number | — | — | 78.3 | — | — | — |
| localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy. | 7.8 | 64.7 | 26.6 | 72.9 | 74.4 | 75.7 |
| regionregionThe first-level administrative subdivision of a country — a US state, a French region, a province. The component between country and locality. | 3.7 | 82.9 | 17.6 | 89.1 | 89.1 | 90.3 |
| postcodepostcodeThe country-specific postal code (US ZIP, French code postal, etc.). Mailwoman handles postcode parsing entirely by rule classifier — a regex problem, not an ML one. | 0.1 | 85.9 | 75.5 | 97.3 | 97.8 | 98.3 |
| country_homograph | — | — | — | 89.8 | 85.1 | 89.8 |
| unitunitA subdivision of a building — apartment, suite, floor — that refines a street address. Mailwoman's unit component; a designator plus identifier forms a subpremise. | — | — | — | 90.6 | 92.1 | 92.1 |
| po_box_real | — | — | — | — | — | 89.1 |
| intersection_real | — | — | — | — | — | 100 |
| cedexCEDEX (Courrier d'Entreprise à Distribution Exceptionnelle). A French postal routing for high-volume business mail: a CEDEX code delivers directly from a sorting centre, bypassing the local post office. A common negative-space format Mailwoman must parse. | — | — | 0 | — | — | — |
| countrycountryThe top-level address component (an ISO country). Closed-vocabulary, so it is best handled by a deterministic matcher feeding a proposal rather than a retrained model head. | 2.5 | 0 | 28 | — | — | — |
| dependent_locality | — | — | 0 | — | — | — |
| venuevenueA named, non-address place — a business, building, park, or stadium. Mailwoman's free-text point-of-interest component, added as a Tier 2 fine label. | — | — | 39.4 | — | — | — |
FR
| tag | 0.1.0 | 0.2.0 | 3.0.0 | 4.2.0 | 4.3.0 | 4.4.0 |
|---|---|---|---|---|---|---|
| house_number | — | — | — | 94.6 | 97.7 | 97.2 |
| regionregionThe first-level administrative subdivision of a country — a US state, a French region, a province. The component between country and locality. | — | — | — | 27.6 | 16.2 | 25.6 |
| postcodepostcodeThe country-specific postal code (US ZIP, French code postal, etc.). Mailwoman handles postcode parsing entirely by rule classifier — a regex problem, not an ML one. | — | — | — | 99.7 | 99.7 | 99.6 |
| cedex_real | — | — | — | — | — | 96.1 |
DE
| tag | 0.1.0 | 0.2.0 | 3.0.0 | 4.2.0 | 4.3.0 | 4.4.0 |
|---|---|---|---|---|---|---|
| native_locality_anchor_on | — | — | — | 90.7 | 90.1 | 91 |