Training and the corpus

The question

The 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.' assigns one of 33 labelscomponent tagOne of the 25 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. to each piece of your input, and it does so because it was shown examples. So: what examples, where do they come from, who labeled them, and what stops a run that improved one thing and broke another from shipping?

The analog

In a rule-based parser you write the rules, and the rules are the specification. Anyone can read them and predict the parser's behavior on an input they invent, because the behavior is the rules.

In a learned parser you write the examples, and the examples are the specification. Everything the parser does on unseen input is an extrapolation from what it was shown — including the extrapolations nobody intended. A 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. that is three-quarters one 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. teaches that 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.'s ordering as a general fact about addresses. A 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. where the localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighborhood in the hierarchy. is nearly always the first tokentokenOne 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. teaches "first tokentokenOne 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., therefore localitylocalityThe city / town / settlement component of an address: a populated place sitting between region and neighborhood in the hierarchy." as a shortcut that beats reading the tokentokenOne 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..

That is the trade. Rules are auditable and bounded; you can read them and you can also read where they end. Examples generalize past what anyone enumerated and they are not auditable the same way — the only way to know what a 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. taught is to test the 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.' on data it did not see.

Everything below is a consequence of taking that trade seriously.

Where labeled rows come from

Nobody hand-labelscomponent tagOne of the 25 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. address text at the scale a 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.' needs. The labelscomponent tagOne of the 25 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. are derived, and the trick is that a large amount of open address data already comes in two halves: a structured record with the components separated into fields, and — or reconstructible from it — the address as a line of text.

National address registersinput modeThe Decision-A register switch: 'fragmented' (human-typed fragments — feeds the evidence channels) vs 'formatted' (complete records — runs the trained absence identity). Explicit on CLI/API; per-endpoint defaults (batch→formatted, autocomplete→fragmented); kind-derived otherwise. publish exactly this. So do census streetstreetThe named linear feature along which house numbers are ordered. Decomposes into a name plus street affixes; one of the Tier 2 fine labels. files, open gazetteersgazetteerA geographical index that maps place names and postcodes to real-world coordinates. Mailwoman uses a custom-built Who's On First (WOF) SQLite database as its gazetteer — the 'atlas' half of the grammar/atlas architecture., and collaborative address projects. Mailwoman's published attribution list names several of them: France's Base Adresse Nationale, Great Britain's Price Paid Data, New Zealand data derived from LINZ, OpenAddressesOpenAddresses (OA). A global open aggregation of address points collected from many official sources. A primary source of component-supervised training data outside proprietary registries. extracts for several European countries, Overture's addresses theme, and Who's On FirstWOF (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. for administrative names. Every row carries the license it arrived under, which is what makes the list an attribution list rather than an inventory.

Turning a structured record into a trainingtrainingThe process of adjusting a model's parameters so its predictions match labeled examples, by repeatedly measuring error and nudging the weights to reduce it. Distinct from inference, when the trained model is run on new input. row is the alignmentalignmentThe step in the corpus pipeline that takes a (raw, components) pair from an adapter and produces a (raw, tokens, BIO labels) row by finding each component's text inside the raw string and labeling the matching tokens. step, and it is more fragile than it sounds. Take the record's components, find each one inside the text line, and mark the first piece it covers B-<tag> and the rest I-<tag>. When a component cannot be located — the record says the regionregionThe first-level administrative subdivision of a country — a US state, a French region, a province. The component between country and locality. is California and the line says CA, or a diacriticdiacriticAn accent mark that modifies a letter (é, ñ, ç). Address normalization must fold diacritics for matching without discarding the information a user typed. differs, or the source's own fields disagree with its own address string — the row is quarantined rather than guessed at. It is written aside with the reason, and it does not train anything.

That quarantine pile is worth more than it looks. A source whose quarantine rate jumps is a source whose schema changed, and it announces itself before the 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.' does.

The multiplier, and the reason for it

Real users do not type addresses the way registersinput modeThe Decision-A register switch: 'fragmented' (human-typed fragments — feeds the evidence channels) vs 'formatted' (complete records — runs the trained absence identity). Explicit on CLI/API; per-endpoint defaults (batch→formatted, autocomplete→fragmented); kind-derived otherwise. store them. They abbreviate 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. type, drop the commas, type in one case, and leave the accents off. A 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. built only from registerinput modeThe Decision-A register switch: 'fragmented' (human-typed fragments — feeds the evidence channels) vs 'formatted' (complete records — runs the trained absence identity). Explicit on CLI/API; per-endpoint defaults (batch→formatted, autocomplete→fragmented); kind-derived otherwise. text teaches the registerinput modeThe Decision-A register switch: 'fragmented' (human-typed fragments — feeds the evidence channels) vs 'formatted' (complete records — runs the trained absence identity). Explicit on CLI/API; per-endpoint defaults (batch→formatted, autocomplete→fragmented); kind-derived otherwise.'s registerinput modeThe Decision-A register switch: 'fragmented' (human-typed fragments — feeds the evidence channels) vs 'formatted' (complete records — runs the trained absence identity). Explicit on CLI/API; per-endpoint defaults (batch→formatted, autocomplete→fragmented); kind-derived otherwise..

So each row is expanded into surface variants before alignmentalignmentThe step in the corpus pipeline that takes a (raw, components) pair from an adapter and produces a (raw, tokens, BIO labels) row by finding each component's text inside the raw string and labeling the matching tokens. runs: shouted and whispered, with the commas and without them, the state name and its abbreviation, Avenue and Ave, Apt and Apartment. Which variants apply is a 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. decision — accent stripping is a French default and does nothing useful for United States rows. Each variant is aligned from scratch rather than having labelscomponent tagOne of the 25 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. copied onto it, because changing the surface moves the boundaries, and a variant that fails to align is quarantined like anything else.

This is why the same idea appears twice in the design. Case restoration at 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. time handles the lowercase input the 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.' was not trained on; case variants at trainingtrainingThe process of adjusting a model's parameters so its predictions match labeled examples, by repeatedly measuring error and nudging the weights to reduce it. Distinct from inference, when the trained model is run on new input. time reduce how much there is to handle. Neither one alone covers the registerinput modeThe Decision-A register switch: 'fragmented' (human-typed fragments — feeds the evidence channels) vs 'formatted' (complete records — runs the trained absence identity). Explicit on CLI/API; per-endpoint defaults (batch→formatted, autocomplete→fragmented); kind-derived otherwise. users type.

What a run optimizes

One thing: for each piece, put as much score as possible on the correct labelcomponent tagOne of the 25 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.. That is per-position cross-entropycross-entropyThe standard classification loss: it penalizes a model for putting low probability on the correct label. Per-token negative log-likelihood is the cross-entropy of each token's label., and it is the whole lossloss functionA number measuring how wrong the model's predictions are on a batch of examples. Training minimizes it. Mailwoman's loss combines per-token negative log-likelihood with the CRF sequence loss.. There is no sequence-level term — no reward for a well-formed reading, no penalty for an impossible transition, because impossibility is enforced at decode time instead. See Decoding and the best path for what was tried on that front and why it is not what ships.

Two levers shape what that lossloss functionA number measuring how wrong the model's predictions are on a batch of examples. Training minimizes it. Mailwoman's loss combines per-token negative log-likelihood with the CRF sequence loss. rewards in practice.

Source weightingsource weightingMultiplicative per-source weights applied during training to oversample underrepresented corpus sources, balancing a corpus dominated by a few large datasets.. Sources differ in size by orders of magnitude, so trainingtrainingThe process of adjusting a model's parameters so its predictions match labeled examples, by repeatedly measuring error and nudging the weights to reduce it. Distinct from inference, when the trained model is run on new input. on the raw mixture lets the largest source define what an address looks like. Weighting lets an under-represented source be drawn more than its row count would suggest. The failure this exists to prevent is positional: a 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. dominated by one source's field order teaches the order rather than the components.

CoveragecoverageThe fraction of a population or region for which a data source has real, non-placeholder entries — e.g. 47% rooftop coverage on Texas addresses. Distinct from accuracy on the rows that are present. of what is missing. A sequence labeler learns a tag partly by learning where it stops, which means teaching it a tag it never had sharpens the tags on either side. Before unit was a category the 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.' could reach for, Apt 456 had to land somewhere, and it landed on street — so street was absorbing anything on the road line and had no reliable end. Giving those piecesECE (Expected Calibration Error). A metric that measures how well a model's confidence scores align with its actual accuracy. Lower is better. Mailwoman's held-out ECE drops from 0.067 (raw) to 0.0035 (calibrated). their own home moved them off street, and street got sharper as a side effect. Adding a category is not free in general — two categories that overlap in the data trade one confusion for another — but where the old labelcomponent tagOne of the 25 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. was swallowing a distinct thing, it is close to it.

Held-out means held-out

The evaluationevalRunning the model against a held-out golden dataset and computing per-component F1, exact-match, calibration, and resolved-coordinate error. portion of the 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. is not a random sample of rows. It is a set of whole regionsregionThe first-level administrative subdivision of a country — a US state, a French region, a province. The component between country and locality., chosen for low data density and lifted out of trainingtrainingThe process of adjusting a model's parameters so its predictions match labeled examples, by repeatedly measuring error and nudging the weights to reduce it. Distinct from inference, when the trained model is run on new input. entirely: Vermont, Wyoming and North Dakota for the United States, Corse, Lozère and Creuse for France, Saarland and Mecklenburg-Vorpommern for Germany. Rows from those regionsregionThe first-level administrative subdivision of a country — a US state, a French region, a province. The component between country and locality. are divided between validation and test by hashing each row's source identifier, and everything else trains.

Split at random instead and the arithmetic flatters you, because addresses leak by neighborhood. 13 Main St, Springfield, IL in trainingtrainingThe process of adjusting a model's parameters so its predictions match labeled examples, by repeatedly measuring error and nudging the weights to reduce it. Distinct from inference, when the trained model is run on new input. and 15 Main St, Springfield, IL in evaluationevalRunning the model against a held-out golden dataset and computing per-component F1, exact-match, calibration, and resolved-coordinate error. scores well by recalling Springfield, not by reading the address — and a 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.' can look like it generalizes while doing nothing but remembering place namestoponymA proper name for a geographic place.. Holding out the whole regionregionThe first-level administrative subdivision of a country — a US state, a French region, a province. The component between country and locality. closes that path, and it makes the measurement answer the question a deployment asks on most of its input: what happens on a place the 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.' has never met.

One rule goes with it, and it is the one that keeps the discipline from decaying. A holdout added after a 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. was built is not a holdout. Every 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. version carries its split as a committed manifest, so a rerun reproduces the same division and nobody can decide after the fact which regionsregionThe first-level administrative subdivision of a country — a US state, a French region, a province. The component between country and locality. were meant to be unseen.

Why an evaluation stands between a run and a release

A trainingtrainingThe process of adjusting a model's parameters so its predictions match labeled examples, by repeatedly measuring error and nudging the weights to reduce it. Distinct from inference, when the trained model is run on new input. run produces a checkpointcheckpointA saved snapshot of the model weights and optimizer state during training. Mailwoman saves a checkpoint periodically so training can resume after a GPU hang. and a number. Neither is a reason to ship.

The gate is per tag and per localelocaleThe combination of language and country an address comes from. en-US and fr-FR are the locales Mailwoman ships weights for., not aggregate. An aggregate score moves when one large category moves, and it is unmoved when a small one collapses — which means shipping on an aggregate is shipping without having looked at the categories the aggregate is made of. First-class localeslocaleThe combination of language and country an address comes from. en-US and fr-FR are the locales Mailwoman ships weights for. carry per-tag floors that block a release when a tag falls below them; other localeslocaleThe combination of language and country an address comes from. en-US and fr-FR are the locales Mailwoman ships weights for. are measured, recorded and argued about rather than gating. Locales and tiers states which localeslocaleThe combination of language and country an address comes from. en-US and fr-FR are the locales Mailwoman ships weights for. are in which tier and what the tier means, and the distinction it draws is the one that matters here: a localelocaleThe combination of language and country an address comes from. en-US and fr-FR are the locales Mailwoman ships weights for. is claimed when there is a graded evaluationevalRunning the model against a held-out golden dataset and computing per-component F1, exact-match, calibration, and resolved-coordinate error. behind it, not when there is trainingtrainingThe process of adjusting a model's parameters so its predictions match labeled examples, by repeatedly measuring error and nudging the weights to reduce it. Distinct from inference, when the trained model is run on new input. data for it.

Two habits go with the gate.

Every shipped 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.' carries its own numbers. The model cardmodel cardA JSON metadata file (model-card.json) shipped with each weights bundle. It declares the model version, lineage, label set, required inference channels (anchor, gazetteer), calibration data, and training provenance. inside the weightsparameterA single learned number inside a model — one weight or bias. Mailwoman's encoder has roughly 30 million of them; training is the search for good values. package publishes the label setlabel setA model's BIO vocabulary, per-config since CJK Phase 2: stage3 (the Latin 33), stage3-jp (STAGE3 + the seven JP tags = 47, the JP char head), stage4 (the secondary-address family — numerically also 47, a coincidence). A checkpoint persists its own id→label map; label-space mismatches raise instead of collapsing silently., the architecture, the calibration measurements, the data attribution, and per-tag scores by localelocaleThe combination of language and country an address comes from. en-US and fr-FR are the locales Mailwoman ships weights for.. It ships to the registry with the 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.', so the claim and the artifact cannot drift apart — you can read what the 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.' you installed was measured at, from the package you installed.

Numbers are recorded per version, not overwritten. A per-version ledger is what makes a regression detectable at all: without the previous version's numbers, measured the same way, "better" has no referent.

What it costs

The 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. decides the competence, and it is uneven. CoveragecoverageThe fraction of a population or region for which a data source has real, non-placeholder entries — e.g. 47% rooftop coverage on Texas addresses. Distinct from accuracy on the rows that are present. follows open data, which follows national policy, which has nothing to do with where your users are. A 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. with a published address registerinput modeThe Decision-A register switch: 'fragmented' (human-typed fragments — feeds the evidence channels) vs 'formatted' (complete records — runs the trained absence identity). Explicit on CLI/API; per-endpoint defaults (batch→formatted, autocomplete→fragmented); kind-derived otherwise. is a 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. the 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.' reads well. A 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. without one is thin no matter how much compute goes into trainingtrainingThe process of adjusting a model's parameters so its predictions match labeled examples, by repeatedly measuring error and nudging the weights to reduce it. Distinct from inference, when the trained model is run on new input..

A source's bias becomes a prior. Every registerinput modeThe Decision-A register switch: 'fragmented' (human-typed fragments — feeds the evidence channels) vs 'formatted' (complete records — runs the trained absence identity). Explicit on CLI/API; per-endpoint defaults (batch→formatted, autocomplete→fragmented); kind-derived otherwise. encodes conventions — which fields are populated, how the regionregionThe first-level administrative subdivision of a country — a US state, a French region, a province. The component between country and locality. is spelled, whether the 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. leads or trails — and those conventions become expectations the 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.' brings to inputs that do not share them.

Improvement is not monotonic. Adding a localelocaleThe combination of language and country an address comes from. en-US and fr-FR are the locales Mailwoman ships weights for. can interfere with an existing one; adding a category can trade one confusion for another. This is the reason the gate is per tag and per localelocaleThe combination of language and country an address comes from. en-US and fr-FR are the locales Mailwoman ships weights for. rather than a single score with a threshold on it, and the reason a run that improves the headline number can still be the wrong thing to ship.

A small evaluationevalRunning the model against a held-out golden dataset and computing per-component F1, exact-match, calibration, and resolved-coordinate error. set cannot tell a small regression from noise. Human-labeled evaluationevalRunning the model against a held-out golden dataset and computing per-component F1, exact-match, calibration, and resolved-coordinate error. data carries annotation error of its own, so a shift of a point or two is inside the noise band and has to be inspected case by case rather than read off the table.

Related

• Tokens and labels — the labelcomponent tagOne of the 25 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. space alignmentalignmentThe step in the corpus pipeline that takes a (raw, components) pair from an adapter and produces a (raw, tokens, BIO labels) row by finding each component's text inside the raw string and labeling the matching tokens. produces.
• Decoding and the best path — what the lossloss functionA number measuring how wrong the model's predictions are on a batch of examples. Training minimizes it. Mailwoman's loss combines per-token negative log-likelihood with the CRF sequence loss. deliberately leaves to decode time.
• What the model cannot do — where the 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. runs out.
• Locales and tiers — what a claimed localelocaleThe combination of language and country an address comes from. en-US and fr-FR are the locales Mailwoman ships weights for. means, and which localeslocaleThe combination of language and country an address comes from. en-US and fr-FR are the locales Mailwoman ships weights for. are claimed.
• What ships today — versions, coveragecoverageThe fraction of a population or region for which a data source has real, non-placeholder entries — e.g. 47% rooftop coverage on Texas addresses. Distinct from accuracy on the rows that are present., and the current gaps.