Skip to main content

Parser failure report

CorpuscorpusThe BIO-labeled training data used to train Mailwoman's neural classifier. Assembled from real sources (OpenAddresses, National Address Database) and synthetic shards (boundary stress, order variants, negative space). Managed by @mailwoman/corpus. parity-corpus.triaged.jsonl · 321 fixtures · modelsneural 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.' shipped, v257, v261, v280. Every gold labelground truthThe correct answer for an example, used as the standard a prediction is graded against. Mailwoman's ground truth is the hand-labeled golden set; its quality caps achievable accuracy. is graded (floor labelscomponent tagOne of the 33 labels in Mailwoman's address schema — street, locality, region, postcode, house_number, unit, po_box, country, venue, intersection, and others. Each parsed span carries exactly one component tag. compare the streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels./prefix/suffix family); a fixture "fails" a labelcomponent tagOne of the 33 labels in Mailwoman's address schema — street, locality, region, postcode, house_number, unit, po_box, country, venue, intersection, and others. Each parsed span carries exactly one component tag. when the parseaddress parsingThe process of decomposing a free-text postal address string into structured components — house number, street name, locality, region, postcode, and country — so a geocoder can resolve them to coordinates. disagrees with the hand gold.

Per-model failure count

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.'fixtures with ≥1 labelcomponent tagOne of the 33 labels in Mailwoman's address schema — street, locality, region, postcode, house_number, unit, po_box, country, venue, intersection, and others. Each parsed span carries exactly one component tag. failurerate
shipped20464%
v25718959%
v26119160%
v28019360%

Failures by label — the class each candidate trades

A count that rises across candidates (bold, with the delta vs the first 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.') is a class the candidate silently traded. This is the row that flags a regression the floor gates miss — e.g. country degrading across the fragment lineage.

labelcomponent tagOne of the 33 labels in Mailwoman's address schema — street, locality, region, postcode, house_number, unit, po_box, country, venue, intersection, and others. Each parsed span carries exactly one component tag.shippedv257v261v280
street150124126127
locality5659 (+3)60 (+4)57
house_number43343436
venue22222222
region15161616
unit15151515
country2211
postcode1111

Correlation — failure rate by structural class

Read across a row: a class that fails more on one 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.' is the shape of what that candidate traded (delimiter-free, non-ASCII, short inputs, per 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.).

structural classshippedv257v261v280
whitespace-only133/193 (69%)122/193 (63%)123/193 (64%)126/193 (65%)
has comma57/113 (50%)54/113 (48%)55/113 (49%)54/113 (48%)
non-ASCII36/44 (82%)33/44 (75%)33/44 (75%)34/44 (77%)
has 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. gold34/42 (81%)34/42 (81%)34/42 (81%)34/42 (81%)
≤3 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.71/119 (60%)61/119 (51%)62/119 (52%)62/119 (52%)
US58/99 (59%)56/99 (57%)57/99 (58%)56/99 (57%)
FR34/43 (79%)28/43 (65%)28/43 (65%)31/43 (72%)
ZZ (synthetic)47/59 (80%)46/59 (78%)46/59 (78%)46/59 (78%)

Beyond reach — fails on every model (180)

The persistent core: addresses no current candidate parsesaddress parsingThe process of decomposing a free-text postal address string into structured components — house number, street name, locality, region, postcode, and country — so a geocoder can resolve them to coordinates.. Per-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. breakdown (the correlation), then a sample.

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.beyond-reach fixtures
US52
ZZ46
FR28
NL10
AU9
NO7
PT6
DE5
PL3
SE3
CZ2
GB2
NZ2
RO2
BR1
ES1
IN1

Sample (first 40 of 180)

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.inputsource
AU6000, NSW, Australiav1-parity:mailwoman/test/address.aus.test.ts
AUUnit 12/345 Main Stv1-parity:mailwoman/test/address.aus.test.ts
AUU 12 345 Main Stv1-parity:mailwoman/test/address.aus.test.ts
AUApartment 12/345 Main Stv1-parity:mailwoman/test/address.aus.test.ts
AUApt 12/345 Main Stv1-parity:mailwoman/test/address.aus.test.ts
AULot 12/345 Main Stv1-parity:mailwoman/test/address.aus.test.ts
AUU12/345 Main Stv1-parity:mailwoman/test/address.aus.test.ts
AULot 12/345 Illawarra Road Marrickville NSW 2204v1-parity:mailwoman/test/address.aus.test.ts
AULot 2, Burrows Avenue, EDMONDSON PARK, NSW, Australiav1-parity:mailwoman/test/address.aus.test.ts
BRRua Raul Leite Magalhães, 65, Tapiraí - SP, 18180-000, Brazilv1-parity:mailwoman/test/address.bra.test.ts
CZKorunní 810, Prahav1-parity:mailwoman/test/address.cze.test.ts
CZKájovská 68, Český Krumlovv1-parity:mailwoman/test/address.cze.test.ts
DEAm Falkpl. 5, 10437 Berlinv1-parity:mailwoman/test/address.deu.test.ts
DEKaschk Bar, Linienstraße 40 10119 Berlinv1-parity:mailwoman/test/address.deu.test.ts
DEGenter Straße 16a, Munich, Germanyv1-parity:mailwoman/test/address.deu.test.ts
DEKönigsallee Düsseldorfv1-parity:mailwoman/test/address.deu.test.ts
DEEberswalder Straße 100 104v1-parity:mailwoman/test/address.deu.test.ts
ESCarrer d'Aragó 155 08011 Barcelonav1-parity:mailwoman/test/address.esp.test.ts
FRRue Montmartrev1-parity:mailwoman/test/address.fra.test.ts
FRRue de Parisv1-parity:mailwoman/test/address.fra.test.ts
FRRue Saint Annev1-parity:mailwoman/test/address.fra.test.ts
FRAllée Victor Hugov1-parity:mailwoman/test/address.fra.test.ts
FRAvenue Aristide Briandv1-parity:mailwoman/test/address.fra.test.ts
FRRue Henri Barbusse Paris Francev1-parity:mailwoman/test/address.fra.test.ts
FRRue du Général Leclerc Dunkerque Francev1-parity:mailwoman/test/address.fra.test.ts
FRAvenue de Sainte Rose de Limav1-parity:mailwoman/test/address.fra.test.ts
FRRue du Capitaine Galinat Marseille Francev1-parity:mailwoman/test/address.fra.test.ts
FRMery Sur Oisev1-parity:mailwoman/test/address.fra.test.ts
FRMéry Sur Oisev1-parity:mailwoman/test/address.fra.test.ts
FR4 Cité Du Cardinal Lemoine 75005 Parisv1-parity:mailwoman/test/address.fra.test.ts
FRRue de l'Inspecteur Alles Parisv1-parity:mailwoman/test/address.fra.test.ts
FRRue de l'Empereur Julien Parisv1-parity:mailwoman/test/address.fra.test.ts
FRRue de l'Adjudant Réau Parisv1-parity:mailwoman/test/address.fra.test.ts
FR10 Boulevard Saint-Germains Parisv1-parity:mailwoman/test/address.fra.test.ts
FREsplanade de la Libertév1-parity:mailwoman/test/address.fra.test.ts
FREsplanade Méditerranéev1-parity:mailwoman/test/address.fra.test.ts
FRÉcole Paul Valéry Montpellierv1-parity:mailwoman/test/place.fra.test.ts
FRUniversité de Montpellierv1-parity:mailwoman/test/place.fra.test.ts
FRÉcole Jules Vernes Villetaneusev1-parity:mailwoman/test/place.fra.test.ts
FRZAC de la Tuilerie, Villars-les-Dombes, Francev1-parity:mailwoman/test/place.fra.test.ts

Model-specific — where candidates disagree (34)

A ✓ under one 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.' and a failure under another = a fix or a regression between candidates — the diff you track release-over-release.

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.inputshippedv257v261v280
AUEight Mile Plains 4113localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.
AUEight Mile Plains 9999localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.
DEAm Bürgerpark 15-18, 13156, BerlinstreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.Am Bürgerpark 1 · house_number→5-18
DERathausplatzstreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.
DEPlutowegstreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.
DEDorfstrassestreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.
FRRue de la PaixstreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.Rue de
FRBoulevard du Général Charles De GaullestreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.Boulevard
FRBoulevard Charles De GaullestreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.
FRRue Jean Baptiste ClémentstreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.Rue Jean
FR32 Rue Du 4 SeptembrestreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.Rue Du 4
FRRue de l'Amiral Galache, ToulousestreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels. · localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.Rue de l'Amiral Galache
FREsplanade du Géneral de GaullestreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.du Géneral de Gaulle
GB24-28 Broadway Market, LondonstreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.Broadway · localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.Market LondonstreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.Broadway · localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.Market London
NLBosserdijk, HooglandstreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels. · localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.Bosserdijk
NLBloemendaallocalitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.
NOSkredderplassen 20streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.
NOGrense Jakobselv veien 311streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.veien · house_number→house_number→
NZ183 Vista Pakuhouse_number→house_number→
PLUlica Strzelecka 12, Nowy SączlocalitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.wy SączlocalitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.wy SączlocalitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.wy Sącz
PLŻorska 11, 47-400streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.Żorska 1 · house_number→1streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.Żorska 1 · house_number→1streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.Żorska 1 · house_number→1
PTTv. da Horta 27AstreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.da Horta · house_number→
ROBulevardul Iuliu Maniu, BuchareststreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels. · localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.Bulevardul Iuliu Maniu
SKDivadelná 41/3, TrnavastreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.Divadeln 4 · house_number→1/3
USwrigley fieldstreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.Field
USFm 3009 TXstreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.Fm · regionregionThe first-level administrative subdivision of a country — a US state, a French region, a province. The component between country and locality.streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.Fm · regionregionThe first-level administrative subdivision of a country — a US state, a French region, a province. The component between country and locality.streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.Fm · regionregionThe first-level administrative subdivision of a country — a US state, a French region, a province. The component between country and locality.
USN DWIGHT AVE Portland OregonstreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.N DWIGHT AVE Portland · localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.N DWIGHT AVE Portland · localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.N DWIGHT AVE Portland · localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.
USHighway 72streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.
USBroadway, ManhattanstreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels. · localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.Broadway Manhattan
USEast Broadway, ManhattanstreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels. · localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.East Broadway Manhattan
US123 Main St New York, NY 10010streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.Main New York St · localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.Main New York St · localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.Main New York St · localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.
USNew York, NYlocalitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy.
USNew York, New YorklocalitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy. · regionregionThe first-level administrative subdivision of a country — a US state, a French region, a province. The component between country and locality.New York New YorklocalitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy. · regionregionThe first-level administrative subdivision of a country — a US state, a French region, a province. The component between country and locality.New York New YorklocalitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighbourhood in the hierarchy. · regionregionThe first-level administrative subdivision of a country — a US state, a French region, a province. The component between country and locality.New York New York
ZZmain plstreetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels.PL