Coarse-placer — across-11 in-map MISROUTING gate (#244, path to default-on)
Generated by scripts/eval/coarse-placer-inmap-misroute.ts. 2000 in-map rows (200/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. × 10; TW excluded — 0 WOFWOF (Who's On First). An open-source gazetteer of places maintained by Mapzen/whosonfirst. Mailwoman builds a custom SQLite database from WOF GeoJSON repos, extended with postcode data, importance scores, and coincident-role relations. rows). Trailing 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. tokenstokenOne word or subword in the tokenized input. For the neural classifier, tokens come from SentencePiece (subword units); for the rule classifiers, tokens are whitespace- and punctuation-separated words. STRIPPED so the 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. must be inferred. Open-set prior, abstainBelow 0.9, anchorWeight 1, modelneural classifierThe machine learning model at the core of Mailwoman's parser — a transformer encoder (~30M parameters) trained from scratch to do BIO token classification over addresses. It learns the 'grammar' of address formats; the gazetteer supplies the 'atlas.' 4.6.0. Right-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. = most-specific resolved node's spr.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., NO defaultCountry.
Net effect: 58 wins, 0 regressions (0.0% of rows)
Right-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. OFF 1014/2000 (50.7%) → ON 1072/2000 (53.6%).
| 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. | n | OFF right | ON right | wins | regressions |
|---|---|---|---|---|---|
| US | 200 | 82.0% | 95.0% | 26 | 0 |
| FR | 200 | 61.5% | 62.0% | 1 | 0 |
| GB | 200 | 35.5% | 44.0% | 17 | 0 |
| CN | 200 | 73.0% | 73.5% | 1 | 0 |
| NL | 200 | 35.5% | 35.5% | 0 | 0 |
| IT | 200 | 54.0% | 56.5% | 5 | 0 |
| DE | 200 | 40.5% | 40.5% | 0 | 0 |
| JP | 200 | 64.0% | 65.5% | 3 | 0 |
| ES | 200 | 35.5% | 37.0% | 3 | 0 |
| KR | 200 | 25.5% | 26.5% | 2 | 0 |
Regression buckets (gold → wrong-country the prior introduced)
None — the prior introduced no in-map misroutes.
Verdict
PASS — zero misroutes; default-on is clean on this probe.