@mailwoman/phrase-grouper 4.0.0

package/out/group.d.ts

@@ -0,0 +1,32 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `groupPhrases` — Stage 2.7 entry point.
+ *
+ *   Composes per-kind rules over the normalized input + QueryShape and emits one `PhraseProposal` per
+ *   fired rule. Overlapping proposals are expected — the reconciler (Stage 5) picks the best
+ *   non-overlapping subset.
+ *
+ *   See `docs/articles/concepts/the-knowledge-ladder.md` § Phrase grouper for the design rationale,
+ *   and `phrase-grouper/rules.ts` for per-rule documentation.
+ */
+import type { GroupPhrasesOpts, LocaleHint, NormalizedInputLite, PhraseProposal, QueryShapeLike } from "./types.js";
+/**
+ * Synchronous, pure rule-based implementation. The async wrapper matches the pipeline contract.
+ *
+ * Emits overlapping proposals freely — the consumer (Stage 5 reconcile) picks the best
+ * non-overlapping subset under semantic+hierarchical constraints. Confidence is a [0,1] score per
+ * proposal; relative ordering is what matters more than absolute calibration at v0.5.0.
+ *
+ * The `_locale` parameter is reserved for future locale-aware rule packs (Japanese
+ * postcode/honorific patterns, French preposition-bound localities) — currently unused.
+ */
+export declare function groupPhrasesSync(input: NormalizedInputLite, shape: QueryShapeLike, _locale?: LocaleHint, _opts?: GroupPhrasesOpts): PhraseProposal[];
+/**
+ * Async variant matching `RuntimePipelineStages.groupPhrases`. Wraps the sync impl so the pipeline
+ * coordinator can use it as-is.
+ */
+export declare function groupPhrases(input: NormalizedInputLite, shape: QueryShapeLike, locale?: LocaleHint, opts?: GroupPhrasesOpts): Promise<PhraseProposal[]>;
+//# sourceMappingURL=group.d.ts.map

package/out/group.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"group.d.ts","sourceRoot":"","sources":["../group.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAaH,OAAO,KAAK,EAAE,gBAAgB,EAAE,UAAU,EAAE,mBAAmB,EAAE,cAAc,EAAE,cAAc,EAAE,MAAM,YAAY,CAAA;AA0BnH;;;;;;;;;GASG;AACH,wBAAgB,gBAAgB,CAC/B,KAAK,EAAE,mBAAmB,EAC1B,KAAK,EAAE,cAAc,EACrB,OAAO,CAAC,EAAE,UAAU,EACpB,KAAK,GAAE,gBAAqB,GAC1B,cAAc,EAAE,CA4BlB;AAED;;;GAGG;AACH,wBAAsB,YAAY,CACjC,KAAK,EAAE,mBAAmB,EAC1B,KAAK,EAAE,cAAc,EACrB,MAAM,CAAC,EAAE,UAAU,EACnB,IAAI,CAAC,EAAE,gBAAgB,GACrB,OAAO,CAAC,cAAc,EAAE,CAAC,CAE3B"}

package/out/group.js

@@ -0,0 +1,80 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `groupPhrases` — Stage 2.7 entry point.
+ *
+ *   Composes per-kind rules over the normalized input + QueryShape and emits one `PhraseProposal` per
+ *   fired rule. Overlapping proposals are expected — the reconciler (Stage 5) picks the best
+ *   non-overlapping subset.
+ *
+ *   See `docs/articles/concepts/the-knowledge-ladder.md` § Phrase grouper for the design rationale,
+ *   and `phrase-grouper/rules.ts` for per-rule documentation.
+ */
+import { scoreHyphenatedCompound, scoreLocalityPhrase, scoreNumeric, scorePostcode, scoreRegionAbbreviation, scoreStreetPhrase, scoreVenuePhrase, tokenizeSegment, } from "./rules.js";
+/**
+ * Walk every QueryShape segment and emit one `tokens-by-segment` list. Falls back to treating the
+ * whole input as a single segment when QueryShape didn't supply segmentation (e.g. callers wiring
+ * the grouper into a path that bypasses QueryShape).
+ */
+function tokensPerSegment(text, shape) {
+    const segs = shape.segments;
+    if (segs && segs.length > 0) {
+        return segs.map((s, idx) => {
+            const start = s.span?.start ?? 0;
+            const end = s.span?.end ?? text.length;
+            return {
+                tokens: tokenizeSegment(text.slice(start, end), start),
+                isFirst: idx === 0,
+                isLast: idx === segs.length - 1,
+            };
+        });
+    }
+    return [{ tokens: tokenizeSegment(text, 0), isFirst: true, isLast: true }];
+}
+/**
+ * Synchronous, pure rule-based implementation. The async wrapper matches the pipeline contract.
+ *
+ * Emits overlapping proposals freely — the consumer (Stage 5 reconcile) picks the best
+ * non-overlapping subset under semantic+hierarchical constraints. Confidence is a [0,1] score per
+ * proposal; relative ordering is what matters more than absolute calibration at v0.5.0.
+ *
+ * The `_locale` parameter is reserved for future locale-aware rule packs (Japanese
+ * postcode/honorific patterns, French preposition-bound localities) — currently unused.
+ */
+export function groupPhrasesSync(input, shape, _locale, _opts = {}) {
+    const text = input.normalized;
+    if (text.length === 0)
+        return [];
+    const proposals = [];
+    // Postcode rule consumes QueryShape directly (segment-agnostic).
+    proposals.push(...scorePostcode(shape, text));
+    // Per-segment rules.
+    for (const { tokens, isFirst, isLast } of tokensPerSegment(text, shape)) {
+        if (tokens.length === 0)
+            continue;
+        proposals.push(...scoreNumeric(tokens, text));
+        proposals.push(...scoreRegionAbbreviation(tokens, text, isLast));
+        proposals.push(...scoreHyphenatedCompound(tokens, text));
+        proposals.push(...scoreStreetPhrase(tokens, text));
+        proposals.push(...scoreLocalityPhrase(tokens, text, isLast));
+        proposals.push(...scoreVenuePhrase(tokens, text, isFirst));
+    }
+    // Sort: descending confidence, ties broken by span start (left-to-right). Downstream Stage 5
+    // can rely on this ordering for top-k selection without re-sorting.
+    proposals.sort((a, b) => {
+        if (a.confidence !== b.confidence)
+            return b.confidence - a.confidence;
+        return a.span.start - b.span.start;
+    });
+    return proposals;
+}
+/**
+ * Async variant matching `RuntimePipelineStages.groupPhrases`. Wraps the sync impl so the pipeline
+ * coordinator can use it as-is.
+ */
+export async function groupPhrases(input, shape, locale, opts) {
+    return groupPhrasesSync(input, shape, locale, opts);
+}
+//# sourceMappingURL=group.js.map

package/out/group.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"group.js","sourceRoot":"","sources":["../group.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;GAaG;AAEH,OAAO,EACN,uBAAuB,EACvB,mBAAmB,EACnB,YAAY,EACZ,aAAa,EACb,uBAAuB,EACvB,iBAAiB,EACjB,gBAAgB,EAChB,eAAe,GAEf,MAAM,YAAY,CAAA;AAGnB;;;;GAIG;AACH,SAAS,gBAAgB,CACxB,IAAY,EACZ,KAAqB;IAErB,MAAM,IAAI,GAAG,KAAK,CAAC,QAAQ,CAAA;IAC3B,IAAI,IAAI,IAAI,IAAI,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC7B,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,EAAE,EAAE;YAC1B,MAAM,KAAK,GAAG,CAAC,CAAC,IAAI,EAAE,KAAK,IAAI,CAAC,CAAA;YAChC,MAAM,GAAG,GAAG,CAAC,CAAC,IAAI,EAAE,GAAG,IAAI,IAAI,CAAC,MAAM,CAAA;YACtC,OAAO;gBACN,MAAM,EAAE,eAAe,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,GAAG,CAAC,EAAE,KAAK,CAAC;gBACtD,OAAO,EAAE,GAAG,KAAK,CAAC;gBAClB,MAAM,EAAE,GAAG,KAAK,IAAI,CAAC,MAAM,GAAG,CAAC;aAC/B,CAAA;QACF,CAAC,CAAC,CAAA;IACH,CAAC;IACD,OAAO,CAAC,EAAE,MAAM,EAAE,eAAe,CAAC,IAAI,EAAE,CAAC,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC,CAAA;AAC3E,CAAC;AAED;;;;;;;;;GASG;AACH,MAAM,UAAU,gBAAgB,CAC/B,KAA0B,EAC1B,KAAqB,EACrB,OAAoB,EACpB,QAA0B,EAAE;IAE5B,MAAM,IAAI,GAAG,KAAK,CAAC,UAAU,CAAA;IAC7B,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,EAAE,CAAA;IAEhC,MAAM,SAAS,GAAqB,EAAE,CAAA;IAEtC,iEAAiE;IACjE,SAAS,CAAC,IAAI,CAAC,GAAG,aAAa,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC,CAAA;IAE7C,qBAAqB;IACrB,KAAK,MAAM,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,IAAI,gBAAgB,CAAC,IAAI,EAAE,KAAK,CAAC,EAAE,CAAC;QACzE,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;YAAE,SAAQ;QACjC,SAAS,CAAC,IAAI,CAAC,GAAG,YAAY,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC,CAAA;QAC7C,SAAS,CAAC,IAAI,CAAC,GAAG,uBAAuB,CAAC,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,CAAC,CAAA;QAChE,SAAS,CAAC,IAAI,CAAC,GAAG,uBAAuB,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC,CAAA;QACxD,SAAS,CAAC,IAAI,CAAC,GAAG,iBAAiB,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC,CAAA;QAClD,SAAS,CAAC,IAAI,CAAC,GAAG,mBAAmB,CAAC,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,CAAC,CAAA;QAC5D,SAAS,CAAC,IAAI,CAAC,GAAG,gBAAgB,CAAC,MAAM,EAAE,IAAI,EAAE,OAAO,CAAC,CAAC,CAAA;IAC3D,CAAC;IAED,6FAA6F;IAC7F,oEAAoE;IACpE,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;QACvB,IAAI,CAAC,CAAC,UAAU,KAAK,CAAC,CAAC,UAAU;YAAE,OAAO,CAAC,CAAC,UAAU,GAAG,CAAC,CAAC,UAAU,CAAA;QACrE,OAAO,CAAC,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,CAAC,IAAI,CAAC,KAAK,CAAA;IACnC,CAAC,CAAC,CAAA;IAEF,OAAO,SAAS,CAAA;AACjB,CAAC;AAED;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,YAAY,CACjC,KAA0B,EAC1B,KAAqB,EACrB,MAAmB,EACnB,IAAuB;IAEvB,OAAO,gBAAgB,CAAC,KAAK,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,CAAC,CAAA;AACpD,CAAC"}

package/out/index.d.ts

@@ -0,0 +1,25 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `@mailwoman/phrase-grouper` — Stage 2.7 of the runtime pipeline.
+ *
+ *   Proposes coherent input units (boundary discovery) with a structural kind hypothesis +
+ *   confidence. Decouples boundary discovery from type classification: Stage 3 conditions on these
+ *   proposals so it answers the simpler "what type is this proposed span?" rather than jointly
+ *   discovering boundaries and types. Stage 5 consumes the proposals as boundary candidates for
+ *   joint decoding.
+ *
+ *   Bitter-lesson-safe: only universal structural cues (proximity, punctuation, capitalization,
+ *   hyphenation, format-shape repetition) — never place-name dictionaries. v0.5.0 ships the
+ *   rule-based v1; learned 1-2M-param span proposer reserved for v0.5.1.
+ *
+ *   See `docs/articles/concepts/the-knowledge-ladder.md` § Phrase grouper for the design rationale
+ *   and `docs/articles/plan/phases/PHASE_8_v0_5_0_fresh_slate.md` § E for the v0.5.0 thread.
+ */
+export { groupPhrases, groupPhrasesSync } from "./group.js";
+export { scoreHyphenatedCompound, scoreLocalityPhrase, scoreNumeric, scorePostcode, scoreRegionAbbreviation, scoreStreetPhrase, scoreVenuePhrase, tokenizeSegment, } from "./rules.js";
+export type { SegmentToken } from "./rules.js";
+export type { GroupPhrasesOpts, LocaleHint, NormalizedInputLite, PhraseGrouper, PhraseKind, PhraseProposal, QueryShapeLike, Section, } from "./types.js";
+//# sourceMappingURL=index.d.ts.map

package/out/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AAEH,OAAO,EAAE,YAAY,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAA;AAC3D,OAAO,EACN,uBAAuB,EACvB,mBAAmB,EACnB,YAAY,EACZ,aAAa,EACb,uBAAuB,EACvB,iBAAiB,EACjB,gBAAgB,EAChB,eAAe,GACf,MAAM,YAAY,CAAA;AACnB,YAAY,EAAE,YAAY,EAAE,MAAM,YAAY,CAAA;AAC9C,YAAY,EACX,gBAAgB,EAChB,UAAU,EACV,mBAAmB,EACnB,aAAa,EACb,UAAU,EACV,cAAc,EACd,cAAc,EACd,OAAO,GACP,MAAM,YAAY,CAAA"}

package/out/index.js

@@ -0,0 +1,23 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   `@mailwoman/phrase-grouper` — Stage 2.7 of the runtime pipeline.
+ *
+ *   Proposes coherent input units (boundary discovery) with a structural kind hypothesis +
+ *   confidence. Decouples boundary discovery from type classification: Stage 3 conditions on these
+ *   proposals so it answers the simpler "what type is this proposed span?" rather than jointly
+ *   discovering boundaries and types. Stage 5 consumes the proposals as boundary candidates for
+ *   joint decoding.
+ *
+ *   Bitter-lesson-safe: only universal structural cues (proximity, punctuation, capitalization,
+ *   hyphenation, format-shape repetition) — never place-name dictionaries. v0.5.0 ships the
+ *   rule-based v1; learned 1-2M-param span proposer reserved for v0.5.1.
+ *
+ *   See `docs/articles/concepts/the-knowledge-ladder.md` § Phrase grouper for the design rationale
+ *   and `docs/articles/plan/phases/PHASE_8_v0_5_0_fresh_slate.md` § E for the v0.5.0 thread.
+ */
+export { groupPhrases, groupPhrasesSync } from "./group.js";
+export { scoreHyphenatedCompound, scoreLocalityPhrase, scoreNumeric, scorePostcode, scoreRegionAbbreviation, scoreStreetPhrase, scoreVenuePhrase, tokenizeSegment, } from "./rules.js";
+//# sourceMappingURL=index.js.map

package/out/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AAEH,OAAO,EAAE,YAAY,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAA;AAC3D,OAAO,EACN,uBAAuB,EACvB,mBAAmB,EACnB,YAAY,EACZ,aAAa,EACb,uBAAuB,EACvB,iBAAiB,EACjB,gBAAgB,EAChB,eAAe,GACf,MAAM,YAAY,CAAA"}

package/out/rules.d.ts

@@ -0,0 +1,91 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   Rule-based scorers for Stage 2.7 phrase grouping. Each rule inspects the tokenized segment +
+ *   QueryShape priors and emits zero or more `PhraseProposal`s with a confidence in [0, 1].
+ *
+ *   Bitter-lesson-safe: only universal structural cues (proximity, punctuation, capitalization,
+ *   hyphenation, format-shape repetition). No place-name dictionaries — a `LOCALITY_PHRASE`
+ *   proposal means "this looks shaped like a multi-word capitalized run that COULD be a city name",
+ *   not "this IS a city name". Typing the span is the classifier's job; this layer only answers "do
+ *   these tokens belong together?".
+ *
+ *   Per "possibilities not constraints", rules emit overlapping proposals freely. The reconciler
+ *   (Stage 5) picks the best non-overlapping subset.
+ */
+import type { PhraseProposal, QueryShapeLike } from "./types.js";
+/**
+ * One token within a segment — absolute offsets into the normalized input. Built by
+ * `tokenizeSegment` from a (segment-text, segment-start) pair.
+ */
+export interface SegmentToken {
+    body: string;
+    start: number;
+    end: number;
+}
+/**
+ * Split a segment body into whitespace-separated tokens. Offsets are absolute into the original
+ * input (caller supplies the segment's `start` offset).
+ */
+export declare function tokenizeSegment(segmentBody: string, segmentStart: number): SegmentToken[];
+/**
+ * `NUMERIC` rule: emit one proposal per all-digit token. House numbers, postcodes (when no format
+ * hit), unit numbers all surface here as a base hypothesis.
+ *
+ * Confidence drops for very long runs (5+ digits) where POSTCODE will typically win; the reconciler
+ * does the final pick.
+ */
+export declare function scoreNumeric(tokens: ReadonlyArray<SegmentToken>, text: string): PhraseProposal[];
+/**
+ * `POSTCODE` rule: lift each `QueryShape.knownFormats` postcode hit directly. The QueryShape stage
+ * already did the format-shape recognition — Stage 2.7's job is just to publish the spans as phrase
+ * proposals so the reconciler can use them.
+ */
+export declare function scorePostcode(shape: QueryShapeLike, text: string): PhraseProposal[];
+/**
+ * `REGION_ABBREVIATION` rule: 2-3 uppercase Latin letters. Tail-of-segment position boosts
+ * confidence because that's the canonical "City, ST ZIP" shape.
+ */
+export declare function scoreRegionAbbreviation(tokens: ReadonlyArray<SegmentToken>, text: string, segmentIsLast: boolean): PhraseProposal[];
+/**
+ * `HYPHENATED_COMPOUND` rule: tokens containing an internal hyphen. Captures `NY-NY` (venue
+ * disambiguation case), `Saint-Denis` (French locality compound), `10118-1234` (ZIP+4 written as a
+ * single token).
+ *
+ * Internal hyphen is the cue; the rule doesn't pre-judge what the compound MEANS — that's typing
+ * (classifier) or reconcile work. A high confidence here just says "this is one unit, not two".
+ */
+export declare function scoreHyphenatedCompound(tokens: ReadonlyArray<SegmentToken>, text: string): PhraseProposal[];
+/**
+ * `STREET_PHRASE` rule: a token run that contains a street-type suffix. The span covers a leading
+ * numeric (house number) when present, through the suffix token.
+ *
+ * Confidence reflects how canonical the run looks: NUMERIC + 1-3 capitalized words + SUFFIX scores
+ * highest; suffix-only or non-leading-numeric variants score lower but still emit.
+ */
+export declare function scoreStreetPhrase(tokens: ReadonlyArray<SegmentToken>, text: string): PhraseProposal[];
+/**
+ * `LOCALITY_PHRASE` rule: runs of contiguous capitalized words (1-4 long). Emits multiple
+ * overlapping proposals so the reconciler can choose between e.g. `Saint Petersburg` as one phrase
+ * vs `Saint` + `Petersburg` as two.
+ *
+ * Confidence scales with: run length (2-3 best), tail-of-segment position, and whether the
+ * preceding token is a comma or segment boundary.
+ */
+export declare function scoreLocalityPhrase(tokens: ReadonlyArray<SegmentToken>, text: string, segmentIsLast: boolean): PhraseProposal[];
+/**
+ * `VENUE_PHRASE` rule: capitalized run containing a venue-marker noun (Steakhouse, Hotel, etc.) OR
+ * containing a hyphenated compound + ≥1 capitalized word.
+ *
+ * The shape "NY-NY Steakhouse" — the kryptonite case the reconciler eventually needs to lift the NY
+ * tokens off REGION — surfaces here as a `VENUE_PHRASE` proposal at moderate-high confidence.
+ *
+ * Also includes a venue-by-exclusion positional prior: multi-word capitalized run in the first
+ * segment with no street suffix, no house number, and no unit marker → weak VENUE_PHRASE at
+ * 0.50-0.55. The idea: if we can't identify what something IS, but it's in the venue slot (first
+ * segment) and doesn't look like any other component, it might be a venue name.
+ */
+export declare function scoreVenuePhrase(tokens: ReadonlyArray<SegmentToken>, text: string, segmentIsFirst?: boolean): PhraseProposal[];
+//# sourceMappingURL=rules.d.ts.map

package/out/rules.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"rules.d.ts","sourceRoot":"","sources":["../rules.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAGH,OAAO,KAAK,EAAE,cAAc,EAAE,cAAc,EAAE,MAAM,YAAY,CAAA;AAEhE;;;GAGG;AACH,MAAM,WAAW,YAAY;IAC5B,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,MAAM,CAAA;IACb,GAAG,EAAE,MAAM,CAAA;CACX;AA+CD;;;GAGG;AACH,wBAAgB,eAAe,CAAC,WAAW,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,GAAG,YAAY,EAAE,CAezF;AAmLD;;;;;;GAMG;AACH,wBAAgB,YAAY,CAAC,MAAM,EAAE,aAAa,CAAC,YAAY,CAAC,EAAE,IAAI,EAAE,MAAM,GAAG,cAAc,EAAE,CAehG;AAED;;;;GAIG;AACH,wBAAgB,aAAa,CAAC,KAAK,EAAE,cAAc,EAAE,IAAI,EAAE,MAAM,GAAG,cAAc,EAAE,CAcnF;AAED;;;GAGG;AACH,wBAAgB,uBAAuB,CACtC,MAAM,EAAE,aAAa,CAAC,YAAY,CAAC,EACnC,IAAI,EAAE,MAAM,EACZ,aAAa,EAAE,OAAO,GACpB,cAAc,EAAE,CAiBlB;AAED;;;;;;;GAOG;AACH,wBAAgB,uBAAuB,CAAC,MAAM,EAAE,aAAa,CAAC,YAAY,CAAC,EAAE,IAAI,EAAE,MAAM,GAAG,cAAc,EAAE,CAc3G;AAED;;;;;;GAMG;AACH,wBAAgB,iBAAiB,CAAC,MAAM,EAAE,aAAa,CAAC,YAAY,CAAC,EAAE,IAAI,EAAE,MAAM,GAAG,cAAc,EAAE,CA+BrG;AAED;;;;;;;GAOG;AACH,wBAAgB,mBAAmB,CAClC,MAAM,EAAE,aAAa,CAAC,YAAY,CAAC,EACnC,IAAI,EAAE,MAAM,EACZ,aAAa,EAAE,OAAO,GACpB,cAAc,EAAE,CA2ClB;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,gBAAgB,CAC/B,MAAM,EAAE,aAAa,CAAC,YAAY,CAAC,EACnC,IAAI,EAAE,MAAM,EACZ,cAAc,CAAC,EAAE,OAAO,GACtB,cAAc,EAAE,CA2ClB"}

package/out/rules.js

@@ -0,0 +1,498 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ *
+ *   Rule-based scorers for Stage 2.7 phrase grouping. Each rule inspects the tokenized segment +
+ *   QueryShape priors and emits zero or more `PhraseProposal`s with a confidence in [0, 1].
+ *
+ *   Bitter-lesson-safe: only universal structural cues (proximity, punctuation, capitalization,
+ *   hyphenation, format-shape repetition). No place-name dictionaries — a `LOCALITY_PHRASE`
+ *   proposal means "this looks shaped like a multi-word capitalized run that COULD be a city name",
+ *   not "this IS a city name". Typing the span is the classifier's job; this layer only answers "do
+ *   these tokens belong together?".
+ *
+ *   Per "possibilities not constraints", rules emit overlapping proposals freely. The reconciler
+ *   (Stage 5) picks the best non-overlapping subset.
+ */
+import { Span } from "@mailwoman/core/tokenization";
+const WHITESPACE = /\s+/;
+const US_REGION_NAMES = new Set([
+    "alabama",
+    "alaska",
+    "arizona",
+    "arkansas",
+    "california",
+    "colorado",
+    "connecticut",
+    "delaware",
+    "florida",
+    "georgia",
+    "hawaii",
+    "idaho",
+    "illinois",
+    "indiana",
+    "iowa",
+    "kansas",
+    "kentucky",
+    "louisiana",
+    "maine",
+    "maryland",
+    "massachusetts",
+    "michigan",
+    "minnesota",
+    "mississippi",
+    "missouri",
+    "montana",
+    "nebraska",
+    "nevada",
+    "ohio",
+    "oklahoma",
+    "oregon",
+    "pennsylvania",
+    "tennessee",
+    "texas",
+    "utah",
+    "vermont",
+    "virginia",
+    "washington",
+    "wisconsin",
+    "wyoming",
+]);
+/**
+ * Split a segment body into whitespace-separated tokens. Offsets are absolute into the original
+ * input (caller supplies the segment's `start` offset).
+ */
+export function tokenizeSegment(segmentBody, segmentStart) {
+    const tokens = [];
+    let i = 0;
+    while (i < segmentBody.length) {
+        while (i < segmentBody.length && WHITESPACE.test(segmentBody[i]))
+            i++;
+        if (i >= segmentBody.length)
+            break;
+        const start = i;
+        while (i < segmentBody.length && !WHITESPACE.test(segmentBody[i]))
+            i++;
+        tokens.push({
+            body: segmentBody.slice(start, i),
+            start: segmentStart + start,
+            end: segmentStart + i,
+        });
+    }
+    return tokens;
+}
+/** Build a `Section` (Span instance) from absolute offsets into the original text. */
+function makeSection(text, start, end) {
+    return Span.from(text.slice(start, end), { start });
+}
+/** True when token body is non-empty digits only. */
+function isAllDigit(s) {
+    return s.length > 0 && /^[0-9]+$/.test(s);
+}
+/** True when token body is 2-3 uppercase Latin letters (US state, Canadian province abbreviation). */
+function isRegionAbbreviation(s) {
+    return /^[A-Z]{2,3}$/.test(s);
+}
+/** True when token starts with an uppercase letter — common Western proper-noun shape. */
+function startsCapitalized(s) {
+    return /^[A-Z]/.test(s);
+}
+/**
+ * Common street-type suffixes (en-US + en-GB + abbreviated forms). Match case-insensitively against
+ * the raw token body. The set is intentionally short — coverage extension belongs in a future
+ * per-locale rule pack, not as a 500-entry dictionary in this rule.
+ */
+const STREET_SUFFIXES = new Set([
+    "st",
+    "st.",
+    "street",
+    "ave",
+    "ave.",
+    "avenue",
+    "blvd",
+    "blvd.",
+    "boulevard",
+    "rd",
+    "rd.",
+    "road",
+    "ln",
+    "ln.",
+    "lane",
+    "dr",
+    "dr.",
+    "drive",
+    "way",
+    "pl",
+    "pl.",
+    "place",
+    "ct",
+    "ct.",
+    "court",
+    "pkwy",
+    "parkway",
+    "hwy",
+    "highway",
+    "ter",
+    "terrace",
+    "cir",
+    "circle",
+    "sq",
+    "square",
+    "trl",
+    "trail",
+]);
+function isStreetSuffix(token) {
+    return STREET_SUFFIXES.has(token.toLowerCase());
+}
+/**
+ * Venue-marker nouns with per-term confidence weights. Same caveat as STREET_SUFFIXES — universal
+ * structural markers, not a places dictionary. Higher weight = stronger venue signal.
+ */
+const VENUE_MARKERS = new Map([
+    // Dining (0.90 — unambiguous venue markers)
+    ["steakhouse", 0.9],
+    ["restaurant", 0.9],
+    ["bistro", 0.9],
+    ["diner", 0.85],
+    ["cafe", 0.85],
+    ["café", 0.85],
+    ["grill", 0.8],
+    ["pizzeria", 0.9],
+    ["bakery", 0.85],
+    ["brewery", 0.85],
+    ["winery", 0.85],
+    ["tavern", 0.8],
+    ["pub", 0.75],
+    ["bar", 0.7],
+    // Lodging
+    ["hotel", 0.9],
+    ["motel", 0.9],
+    ["inn", 0.75],
+    ["resort", 0.85],
+    ["lodge", 0.75],
+    ["hostel", 0.85],
+    // Entertainment / culture
+    ["theater", 0.85],
+    ["theatre", 0.85],
+    ["cinema", 0.85],
+    ["stadium", 0.9],
+    ["arena", 0.85],
+    ["museum", 0.85],
+    ["gallery", 0.75],
+    ["casino", 0.85],
+    ["lounge", 0.7],
+    // Retail / commercial
+    ["market", 0.7],
+    ["mall", 0.8],
+    ["plaza", 0.7],
+    ["tower", 0.65],
+    ["center", 0.6],
+    ["centre", 0.6],
+    // Medical / institutional
+    ["hospital", 0.9],
+    ["clinic", 0.85],
+    ["pharmacy", 0.85],
+    // Education
+    ["university", 0.9],
+    ["college", 0.85],
+    ["school", 0.8],
+    ["academy", 0.8],
+    // Civic / religious
+    ["church", 0.8],
+    ["temple", 0.8],
+    ["mosque", 0.8],
+    ["synagogue", 0.85],
+    ["cathedral", 0.85],
+    ["chapel", 0.75],
+    ["library", 0.85],
+    // Outdoor
+    ["park", 0.6],
+    ["gardens", 0.65],
+    ["ranch", 0.7],
+    ["farm", 0.65],
+]);
+/**
+ * Unit-designator tokens that gate the venue-by-exclusion heuristic. When any token in a segment
+ * matches one of these, the segment is likely a unit/suite line, not a venue name.
+ */
+const UNIT_MARKERS = new Set([
+    "apt",
+    "apt.",
+    "apartment",
+    "unit",
+    "ste",
+    "ste.",
+    "suite",
+    "room",
+    "rm",
+    "rm.",
+    "floor",
+    "fl",
+    "fl.",
+    "bldg",
+    "bldg.",
+    "building",
+    "dept",
+    "dept.",
+    "department",
+    "#",
+]);
+function venueMarkerWeight(tokens) {
+    let maxWeight = 0;
+    for (const t of tokens) {
+        const w = VENUE_MARKERS.get(t.body.toLowerCase());
+        if (w !== undefined && w > maxWeight)
+            maxWeight = w;
+    }
+    return maxWeight;
+}
+function hasUnitMarker(tokens) {
+    return tokens.some((t) => UNIT_MARKERS.has(t.body.toLowerCase()));
+}
+/**
+ * `NUMERIC` rule: emit one proposal per all-digit token. House numbers, postcodes (when no format
+ * hit), unit numbers all surface here as a base hypothesis.
+ *
+ * Confidence drops for very long runs (5+ digits) where POSTCODE will typically win; the reconciler
+ * does the final pick.
+ */
+export function scoreNumeric(tokens, text) {
+    const out = [];
+    for (const t of tokens) {
+        if (!isAllDigit(t.body))
+            continue;
+        const len = t.body.length;
+        // 1-4 digit pure-numerics are clearly NUMERIC (house number). 5+ are ambiguous with POSTCODE
+        // — emit anyway at lower confidence so the reconciler sees both options.
+        const confidence = len <= 4 ? 0.95 : 0.55;
+        out.push({
+            span: makeSection(text, t.start, t.end),
+            kindHypothesis: "NUMERIC",
+            confidence,
+        });
+    }
+    return out;
+}
+/**
+ * `POSTCODE` rule: lift each `QueryShape.knownFormats` postcode hit directly. The QueryShape stage
+ * already did the format-shape recognition — Stage 2.7's job is just to publish the spans as phrase
+ * proposals so the reconciler can use them.
+ */
+export function scorePostcode(shape, text) {
+    const out = [];
+    for (const hit of shape.knownFormats) {
+        // `po_box` is not a postcode; the kind classifier owns that signal. Skip non-postcode
+        // formats here so we don't pollute POSTCODE proposals.
+        if (hit.format === "po_box")
+            continue;
+        out.push({
+            span: makeSection(text, hit.span.start, hit.span.end),
+            kindHypothesis: "POSTCODE",
+            // Lift the format-hit confidence directly — Stage 5 can weight it against alternatives.
+            confidence: hit.confidence,
+        });
+    }
+    return out;
+}
+/**
+ * `REGION_ABBREVIATION` rule: 2-3 uppercase Latin letters. Tail-of-segment position boosts
+ * confidence because that's the canonical "City, ST ZIP" shape.
+ */
+export function scoreRegionAbbreviation(tokens, text, segmentIsLast) {
+    const out = [];
+    for (let i = 0; i < tokens.length; i++) {
+        const t = tokens[i];
+        if (!isRegionAbbreviation(t.body))
+            continue;
+        // Position cue: last token in a segment (canonical region slot) → high confidence. Anywhere
+        // else, moderate. Anywhere in the LAST segment → slightly elevated (region is canonically the
+        // final non-postcode component).
+        const atTail = i === tokens.length - 1;
+        const confidence = atTail ? 0.85 : segmentIsLast ? 0.7 : 0.55;
+        out.push({
+            span: makeSection(text, t.start, t.end),
+            kindHypothesis: "REGION_ABBREVIATION",
+            confidence,
+        });
+    }
+    return out;
+}
+/**
+ * `HYPHENATED_COMPOUND` rule: tokens containing an internal hyphen. Captures `NY-NY` (venue
+ * disambiguation case), `Saint-Denis` (French locality compound), `10118-1234` (ZIP+4 written as a
+ * single token).
+ *
+ * Internal hyphen is the cue; the rule doesn't pre-judge what the compound MEANS — that's typing
+ * (classifier) or reconcile work. A high confidence here just says "this is one unit, not two".
+ */
+export function scoreHyphenatedCompound(tokens, text) {
+    const out = [];
+    for (const t of tokens) {
+        if (!t.body.includes("-"))
+            continue;
+        // Skip leading/trailing hyphens (likely punctuation drift) — require an interior hyphen
+        // surrounded by non-hyphen characters.
+        if (!/[^-]-[^-]/.test(t.body))
+            continue;
+        out.push({
+            span: makeSection(text, t.start, t.end),
+            kindHypothesis: "HYPHENATED_COMPOUND",
+            confidence: 0.88,
+        });
+    }
+    return out;
+}
+/**
+ * `STREET_PHRASE` rule: a token run that contains a street-type suffix. The span covers a leading
+ * numeric (house number) when present, through the suffix token.
+ *
+ * Confidence reflects how canonical the run looks: NUMERIC + 1-3 capitalized words + SUFFIX scores
+ * highest; suffix-only or non-leading-numeric variants score lower but still emit.
+ */
+export function scoreStreetPhrase(tokens, text) {
+    const out = [];
+    for (let suffixIdx = 0; suffixIdx < tokens.length; suffixIdx++) {
+        if (!isStreetSuffix(tokens[suffixIdx].body))
+            continue;
+        // Walk left from the suffix gathering capitalized/numeric/ordinal tokens. Stop when we hit
+        // something un-street-y (lowercase non-suffix, another suffix, etc.).
+        let start = suffixIdx;
+        for (let i = suffixIdx - 1; i >= 0; i--) {
+            const body = tokens[i].body;
+            if (isAllDigit(body) || /^\d+(st|nd|rd|th)$/i.test(body) || startsCapitalized(body)) {
+                start = i;
+            }
+            else {
+                break;
+            }
+        }
+        // Need at least one preceding token (or a numeric house number) for STREET_PHRASE — a
+        // suffix-only token "Street" alone isn't a street phrase.
+        if (start === suffixIdx)
+            continue;
+        const startTok = tokens[start];
+        const endTok = tokens[suffixIdx];
+        const hasLeadingNumeric = isAllDigit(startTok.body) || /^\d+(st|nd|rd|th)$/i.test(startTok.body);
+        // Canonical NUMERIC + capitalized + SUFFIX scores high; capitalized-run + SUFFIX scores
+        // slightly lower since it could also be a venue.
+        const confidence = hasLeadingNumeric ? 0.9 : 0.75;
+        out.push({
+            span: makeSection(text, startTok.start, endTok.end),
+            kindHypothesis: "STREET_PHRASE",
+            confidence,
+        });
+    }
+    return out;
+}
+/**
+ * `LOCALITY_PHRASE` rule: runs of contiguous capitalized words (1-4 long). Emits multiple
+ * overlapping proposals so the reconciler can choose between e.g. `Saint Petersburg` as one phrase
+ * vs `Saint` + `Petersburg` as two.
+ *
+ * Confidence scales with: run length (2-3 best), tail-of-segment position, and whether the
+ * preceding token is a comma or segment boundary.
+ */
+export function scoreLocalityPhrase(tokens, text, segmentIsLast) {
+    const out = [];
+    for (let i = 0; i < tokens.length; i++) {
+        if (!startsCapitalized(tokens[i].body))
+            continue;
+        // Skip pure region abbreviations as standalone LOCALITY_PHRASE — they own REGION_ABBREVIATION
+        // at higher confidence.
+        if (isRegionAbbreviation(tokens[i].body))
+            continue;
+        // Walk forward grabbing additional capitalized tokens. Stop on lowercase, digit-only, or
+        // region-abbreviation tokens.
+        let j = i;
+        while (j + 1 < tokens.length &&
+            startsCapitalized(tokens[j + 1].body) &&
+            !isRegionAbbreviation(tokens[j + 1].body) &&
+            !isAllDigit(tokens[j + 1].body) &&
+            !isStreetSuffix(tokens[j + 1].body)) {
+            j++;
+        }
+        // Emit proposals for every prefix-length of the run starting at i, capped at 4 tokens.
+        // Each starting i contributes at most 4 proposals, so the rule stays O(n) per segment.
+        const maxLen = Math.min(j - i + 1, 4);
+        for (let len = 1; len <= maxLen; len++) {
+            const startTok = tokens[i];
+            const endTok = tokens[i + len - 1];
+            const spanText = text.slice(startTok.start, endTok.end);
+            const isRegionName = len === 1 && US_REGION_NAMES.has(spanText.toLowerCase());
+            const atTail = i + len - 1 === tokens.length - 1;
+            let confidence = 0.55 + (len === 2 ? 0.15 : 0) + (len === 3 ? 0.1 : 0);
+            if (isRegionName && !atTail)
+                confidence -= 0.2;
+            if (atTail && segmentIsLast)
+                confidence += 0.1;
+            if (atTail)
+                confidence += 0.05;
+            out.push({
+                span: makeSection(text, startTok.start, endTok.end),
+                kindHypothesis: "LOCALITY_PHRASE",
+                confidence: Math.min(0.95, confidence),
+            });
+        }
+        // Do NOT skip past the run — let i++ advance normally so every capitalized token gets a
+        // chance to emit single-token proposals from its own starting position. (Saint Petersburg
+        // needs `Saint`, `Petersburg`, AND `Saint Petersburg`; a run-skip would lose `Petersburg`.)
+    }
+    return out;
+}
+/**
+ * `VENUE_PHRASE` rule: capitalized run containing a venue-marker noun (Steakhouse, Hotel, etc.) OR
+ * containing a hyphenated compound + ≥1 capitalized word.
+ *
+ * The shape "NY-NY Steakhouse" — the kryptonite case the reconciler eventually needs to lift the NY
+ * tokens off REGION — surfaces here as a `VENUE_PHRASE` proposal at moderate-high confidence.
+ *
+ * Also includes a venue-by-exclusion positional prior: multi-word capitalized run in the first
+ * segment with no street suffix, no house number, and no unit marker → weak VENUE_PHRASE at
+ * 0.50-0.55. The idea: if we can't identify what something IS, but it's in the venue slot (first
+ * segment) and doesn't look like any other component, it might be a venue name.
+ */
+export function scoreVenuePhrase(tokens, text, segmentIsFirst) {
+    const out = [];
+    let i = 0;
+    while (i < tokens.length) {
+        if (!startsCapitalized(tokens[i].body)) {
+            i++;
+            continue;
+        }
+        let j = i;
+        while (j + 1 < tokens.length && (startsCapitalized(tokens[j + 1].body) || tokens[j + 1].body.includes("-"))) {
+            j++;
+        }
+        const run = tokens.slice(i, j + 1);
+        const markerWeight = venueMarkerWeight(run);
+        const hasHyphenCompound = run.some((t) => /[^-]-[^-]/.test(t.body));
+        if (markerWeight > 0 || (hasHyphenCompound && run.length >= 2)) {
+            const startTok = run[0];
+            const endTok = run[run.length - 1];
+            const confidence = markerWeight > 0 ? markerWeight : 0.65;
+            out.push({
+                span: makeSection(text, startTok.start, endTok.end),
+                kindHypothesis: "VENUE_PHRASE",
+                confidence,
+            });
+        }
+        else if (segmentIsFirst && run.length >= 2) {
+            const hasStreet = run.some((t) => isStreetSuffix(t.body));
+            const hasLeadingNum = isAllDigit(run[0].body);
+            const hasUnit = hasUnitMarker(run);
+            if (!hasStreet && !hasLeadingNum && !hasUnit) {
+                const startTok = run[0];
+                const endTok = run[run.length - 1];
+                out.push({
+                    span: makeSection(text, startTok.start, endTok.end),
+                    kindHypothesis: "VENUE_PHRASE",
+                    confidence: run.length >= 3 ? 0.55 : 0.5,
+                });
+            }
+        }
+        i = j + 1;
+    }
+    return out;
+}
+//# sourceMappingURL=rules.js.map

package/out/rules.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"rules.js","sourceRoot":"","sources":["../rules.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAEH,OAAO,EAAE,IAAI,EAAE,MAAM,8BAA8B,CAAA;AAanD,MAAM,UAAU,GAAG,KAAK,CAAA;AAExB,MAAM,eAAe,GAAwB,IAAI,GAAG,CAAC;IACpD,SAAS;IACT,QAAQ;IACR,SAAS;IACT,UAAU;IACV,YAAY;IACZ,UAAU;IACV,aAAa;IACb,UAAU;IACV,SAAS;IACT,SAAS;IACT,QAAQ;IACR,OAAO;IACP,UAAU;IACV,SAAS;IACT,MAAM;IACN,QAAQ;IACR,UAAU;IACV,WAAW;IACX,OAAO;IACP,UAAU;IACV,eAAe;IACf,UAAU;IACV,WAAW;IACX,aAAa;IACb,UAAU;IACV,SAAS;IACT,UAAU;IACV,QAAQ;IACR,MAAM;IACN,UAAU;IACV,QAAQ;IACR,cAAc;IACd,WAAW;IACX,OAAO;IACP,MAAM;IACN,SAAS;IACT,UAAU;IACV,YAAY;IACZ,WAAW;IACX,SAAS;CACT,CAAC,CAAA;AAEF;;;GAGG;AACH,MAAM,UAAU,eAAe,CAAC,WAAmB,EAAE,YAAoB;IACxE,MAAM,MAAM,GAAmB,EAAE,CAAA;IACjC,IAAI,CAAC,GAAG,CAAC,CAAA;IACT,OAAO,CAAC,GAAG,WAAW,CAAC,MAAM,EAAE,CAAC;QAC/B,OAAO,CAAC,GAAG,WAAW,CAAC,MAAM,IAAI,UAAU,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,CAAE,CAAC;YAAE,CAAC,EAAE,CAAA;QACtE,IAAI,CAAC,IAAI,WAAW,CAAC,MAAM;YAAE,MAAK;QAClC,MAAM,KAAK,GAAG,CAAC,CAAA;QACf,OAAO,CAAC,GAAG,WAAW,CAAC,MAAM,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC,CAAE,CAAC;YAAE,CAAC,EAAE,CAAA;QACvE,MAAM,CAAC,IAAI,CAAC;YACX,IAAI,EAAE,WAAW,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC;YACjC,KAAK,EAAE,YAAY,GAAG,KAAK;YAC3B,GAAG,EAAE,YAAY,GAAG,CAAC;SACrB,CAAC,CAAA;IACH,CAAC;IACD,OAAO,MAAM,CAAA;AACd,CAAC;AAED,sFAAsF;AACtF,SAAS,WAAW,CAAC,IAAY,EAAE,KAAa,EAAE,GAAW;IAC5D,OAAO,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,GAAG,CAAC,EAAE,EAAE,KAAK,EAAE,CAAC,CAAA;AACpD,CAAC;AAED,qDAAqD;AACrD,SAAS,UAAU,CAAC,CAAS;IAC5B,OAAO,CAAC,CAAC,MAAM,GAAG,CAAC,IAAI,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,CAAA;AAC1C,CAAC;AAED,sGAAsG;AACtG,SAAS,oBAAoB,CAAC,CAAS;IACtC,OAAO,cAAc,CAAC,IAAI,CAAC,CAAC,CAAC,CAAA;AAC9B,CAAC;AAED,0FAA0F;AAC1F,SAAS,iBAAiB,CAAC,CAAS;IACnC,OAAO,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,CAAA;AACxB,CAAC;AAED;;;;GAIG;AACH,MAAM,eAAe,GAAwB,IAAI,GAAG,CAAC;IACpD,IAAI;IACJ,KAAK;IACL,QAAQ;IACR,KAAK;IACL,MAAM;IACN,QAAQ;IACR,MAAM;IACN,OAAO;IACP,WAAW;IACX,IAAI;IACJ,KAAK;IACL,MAAM;IACN,IAAI;IACJ,KAAK;IACL,MAAM;IACN,IAAI;IACJ,KAAK;IACL,OAAO;IACP,KAAK;IACL,IAAI;IACJ,KAAK;IACL,OAAO;IACP,IAAI;IACJ,KAAK;IACL,OAAO;IACP,MAAM;IACN,SAAS;IACT,KAAK;IACL,SAAS;IACT,KAAK;IACL,SAAS;IACT,KAAK;IACL,QAAQ;IACR,IAAI;IACJ,QAAQ;IACR,KAAK;IACL,OAAO;CACP,CAAC,CAAA;AAEF,SAAS,cAAc,CAAC,KAAa;IACpC,OAAO,eAAe,CAAC,GAAG,CAAC,KAAK,CAAC,WAAW,EAAE,CAAC,CAAA;AAChD,CAAC;AAED;;;GAGG;AACH,MAAM,aAAa,GAAgC,IAAI,GAAG,CAAC;IAC1D,4CAA4C;IAC5C,CAAC,YAAY,EAAE,GAAG,CAAC;IACnB,CAAC,YAAY,EAAE,GAAG,CAAC;IACnB,CAAC,QAAQ,EAAE,GAAG,CAAC;IACf,CAAC,OAAO,EAAE,IAAI,CAAC;IACf,CAAC,MAAM,EAAE,IAAI,CAAC;IACd,CAAC,MAAM,EAAE,IAAI,CAAC;IACd,CAAC,OAAO,EAAE,GAAG,CAAC;IACd,CAAC,UAAU,EAAE,GAAG,CAAC;IACjB,CAAC,QAAQ,EAAE,IAAI,CAAC;IAChB,CAAC,SAAS,EAAE,IAAI,CAAC;IACjB,CAAC,QAAQ,EAAE,IAAI,CAAC;IAChB,CAAC,QAAQ,EAAE,GAAG,CAAC;IACf,CAAC,KAAK,EAAE,IAAI,CAAC;IACb,CAAC,KAAK,EAAE,GAAG,CAAC;IACZ,UAAU;IACV,CAAC,OAAO,EAAE,GAAG,CAAC;IACd,CAAC,OAAO,EAAE,GAAG,CAAC;IACd,CAAC,KAAK,EAAE,IAAI,CAAC;IACb,CAAC,QAAQ,EAAE,IAAI,CAAC;IAChB,CAAC,OAAO,EAAE,IAAI,CAAC;IACf,CAAC,QAAQ,EAAE,IAAI,CAAC;IAChB,0BAA0B;IAC1B,CAAC,SAAS,EAAE,IAAI,CAAC;IACjB,CAAC,SAAS,EAAE,IAAI,CAAC;IACjB,CAAC,QAAQ,EAAE,IAAI,CAAC;IAChB,CAAC,SAAS,EAAE,GAAG,CAAC;IAChB,CAAC,OAAO,EAAE,IAAI,CAAC;IACf,CAAC,QAAQ,EAAE,IAAI,CAAC;IAChB,CAAC,SAAS,EAAE,IAAI,CAAC;IACjB,CAAC,QAAQ,EAAE,IAAI,CAAC;IAChB,CAAC,QAAQ,EAAE,GAAG,CAAC;IACf,sBAAsB;IACtB,CAAC,QAAQ,EAAE,GAAG,CAAC;IACf,CAAC,MAAM,EAAE,GAAG,CAAC;IACb,CAAC,OAAO,EAAE,GAAG,CAAC;IACd,CAAC,OAAO,EAAE,IAAI,CAAC;IACf,CAAC,QAAQ,EAAE,GAAG,CAAC;IACf,CAAC,QAAQ,EAAE,GAAG,CAAC;IACf,0BAA0B;IAC1B,CAAC,UAAU,EAAE,GAAG,CAAC;IACjB,CAAC,QAAQ,EAAE,IAAI,CAAC;IAChB,CAAC,UAAU,EAAE,IAAI,CAAC;IAClB,YAAY;IACZ,CAAC,YAAY,EAAE,GAAG,CAAC;IACnB,CAAC,SAAS,EAAE,IAAI,CAAC;IACjB,CAAC,QAAQ,EAAE,GAAG,CAAC;IACf,CAAC,SAAS,EAAE,GAAG,CAAC;IAChB,oBAAoB;IACpB,CAAC,QAAQ,EAAE,GAAG,CAAC;IACf,CAAC,QAAQ,EAAE,GAAG,CAAC;IACf,CAAC,QAAQ,EAAE,GAAG,CAAC;IACf,CAAC,WAAW,EAAE,IAAI,CAAC;IACnB,CAAC,WAAW,EAAE,IAAI,CAAC;IACnB,CAAC,QAAQ,EAAE,IAAI,CAAC;IAChB,CAAC,SAAS,EAAE,IAAI,CAAC;IACjB,UAAU;IACV,CAAC,MAAM,EAAE,GAAG,CAAC;IACb,CAAC,SAAS,EAAE,IAAI,CAAC;IACjB,CAAC,OAAO,EAAE,GAAG,CAAC;IACd,CAAC,MAAM,EAAE,IAAI,CAAC;CACd,CAAC,CAAA;AAEF;;;GAGG;AACH,MAAM,YAAY,GAAwB,IAAI,GAAG,CAAC;IACjD,KAAK;IACL,MAAM;IACN,WAAW;IACX,MAAM;IACN,KAAK;IACL,MAAM;IACN,OAAO;IACP,MAAM;IACN,IAAI;IACJ,KAAK;IACL,OAAO;IACP,IAAI;IACJ,KAAK;IACL,MAAM;IACN,OAAO;IACP,UAAU;IACV,MAAM;IACN,OAAO;IACP,YAAY;IACZ,GAAG;CACH,CAAC,CAAA;AAEF,SAAS,iBAAiB,CAAC,MAAmC;IAC7D,IAAI,SAAS,GAAG,CAAC,CAAA;IACjB,KAAK,MAAM,CAAC,IAAI,MAAM,EAAE,CAAC;QACxB,MAAM,CAAC,GAAG,aAAa,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC,CAAA;QACjD,IAAI,CAAC,KAAK,SAAS,IAAI,CAAC,GAAG,SAAS;YAAE,SAAS,GAAG,CAAC,CAAA;IACpD,CAAC;IACD,OAAO,SAAS,CAAA;AACjB,CAAC;AAED,SAAS,aAAa,CAAC,MAAmC;IACzD,OAAO,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,WAAW,EAAE,CAAC,CAAC,CAAA;AAClE,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,YAAY,CAAC,MAAmC,EAAE,IAAY;IAC7E,MAAM,GAAG,GAAqB,EAAE,CAAA;IAChC,KAAK,MAAM,CAAC,IAAI,MAAM,EAAE,CAAC;QACxB,IAAI,CAAC,UAAU,CAAC,CAAC,CAAC,IAAI,CAAC;YAAE,SAAQ;QACjC,MAAM,GAAG,GAAG,CAAC,CAAC,IAAI,CAAC,MAAM,CAAA;QACzB,6FAA6F;QAC7F,yEAAyE;QACzE,MAAM,UAAU,GAAG,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAA;QACzC,GAAG,CAAC,IAAI,CAAC;YACR,IAAI,EAAE,WAAW,CAAC,IAAI,EAAE,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,GAAG,CAAC;YACvC,cAAc,EAAE,SAAS;YACzB,UAAU;SACV,CAAC,CAAA;IACH,CAAC;IACD,OAAO,GAAG,CAAA;AACX,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,aAAa,CAAC,KAAqB,EAAE,IAAY;IAChE,MAAM,GAAG,GAAqB,EAAE,CAAA;IAChC,KAAK,MAAM,GAAG,IAAI,KAAK,CAAC,YAAY,EAAE,CAAC;QACtC,sFAAsF;QACtF,uDAAuD;QACvD,IAAI,GAAG,CAAC,MAAM,KAAK,QAAQ;YAAE,SAAQ;QACrC,GAAG,CAAC,IAAI,CAAC;YACR,IAAI,EAAE,WAAW,CAAC,IAAI,EAAE,GAAG,CAAC,IAAI,CAAC,KAAK,EAAE,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC;YACrD,cAAc,EAAE,UAAU;YAC1B,wFAAwF;YACxF,UAAU,EAAE,GAAG,CAAC,UAAU;SAC1B,CAAC,CAAA;IACH,CAAC;IACD,OAAO,GAAG,CAAA;AACX,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,uBAAuB,CACtC,MAAmC,EACnC,IAAY,EACZ,aAAsB;IAEtB,MAAM,GAAG,GAAqB,EAAE,CAAA;IAChC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACxC,MAAM,CAAC,GAAG,MAAM,CAAC,CAAC,CAAE,CAAA;QACpB,IAAI,CAAC,oBAAoB,CAAC,CAAC,CAAC,IAAI,CAAC;YAAE,SAAQ;QAC3C,4FAA4F;QAC5F,8FAA8F;QAC9F,iCAAiC;QACjC,MAAM,MAAM,GAAG,CAAC,KAAK,MAAM,CAAC,MAAM,GAAG,CAAC,CAAA;QACtC,MAAM,UAAU,GAAG,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAA;QAC7D,GAAG,CAAC,IAAI,CAAC;YACR,IAAI,EAAE,WAAW,CAAC,IAAI,EAAE,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,GAAG,CAAC;YACvC,cAAc,EAAE,qBAAqB;YACrC,UAAU;SACV,CAAC,CAAA;IACH,CAAC;IACD,OAAO,GAAG,CAAA;AACX,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,uBAAuB,CAAC,MAAmC,EAAE,IAAY;IACxF,MAAM,GAAG,GAAqB,EAAE,CAAA;IAChC,KAAK,MAAM,CAAC,IAAI,MAAM,EAAE,CAAC;QACxB,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC;YAAE,SAAQ;QACnC,wFAAwF;QACxF,uCAAuC;QACvC,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC;YAAE,SAAQ;QACvC,GAAG,CAAC,IAAI,CAAC;YACR,IAAI,EAAE,WAAW,CAAC,IAAI,EAAE,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,GAAG,CAAC;YACvC,cAAc,EAAE,qBAAqB;YACrC,UAAU,EAAE,IAAI;SAChB,CAAC,CAAA;IACH,CAAC;IACD,OAAO,GAAG,CAAA;AACX,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,iBAAiB,CAAC,MAAmC,EAAE,IAAY;IAClF,MAAM,GAAG,GAAqB,EAAE,CAAA;IAChC,KAAK,IAAI,SAAS,GAAG,CAAC,EAAE,SAAS,GAAG,MAAM,CAAC,MAAM,EAAE,SAAS,EAAE,EAAE,CAAC;QAChE,IAAI,CAAC,cAAc,CAAC,MAAM,CAAC,SAAS,CAAE,CAAC,IAAI,CAAC;YAAE,SAAQ;QACtD,2FAA2F;QAC3F,sEAAsE;QACtE,IAAI,KAAK,GAAG,SAAS,CAAA;QACrB,KAAK,IAAI,CAAC,GAAG,SAAS,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;YACzC,MAAM,IAAI,GAAG,MAAM,CAAC,CAAC,CAAE,CAAC,IAAI,CAAA;YAC5B,IAAI,UAAU,CAAC,IAAI,CAAC,IAAI,qBAAqB,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,iBAAiB,CAAC,IAAI,CAAC,EAAE,CAAC;gBACrF,KAAK,GAAG,CAAC,CAAA;YACV,CAAC;iBAAM,CAAC;gBACP,MAAK;YACN,CAAC;QACF,CAAC;QACD,sFAAsF;QACtF,0DAA0D;QAC1D,IAAI,KAAK,KAAK,SAAS;YAAE,SAAQ;QACjC,MAAM,QAAQ,GAAG,MAAM,CAAC,KAAK,CAAE,CAAA;QAC/B,MAAM,MAAM,GAAG,MAAM,CAAC,SAAS,CAAE,CAAA;QACjC,MAAM,iBAAiB,GAAG,UAAU,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,qBAAqB,CAAC,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAA;QAChG,wFAAwF;QACxF,iDAAiD;QACjD,MAAM,UAAU,GAAG,iBAAiB,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAA;QACjD,GAAG,CAAC,IAAI,CAAC;YACR,IAAI,EAAE,WAAW,CAAC,IAAI,EAAE,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC,GAAG,CAAC;YACnD,cAAc,EAAE,eAAe;YAC/B,UAAU;SACV,CAAC,CAAA;IACH,CAAC;IACD,OAAO,GAAG,CAAA;AACX,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,mBAAmB,CAClC,MAAmC,EACnC,IAAY,EACZ,aAAsB;IAEtB,MAAM,GAAG,GAAqB,EAAE,CAAA;IAChC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACxC,IAAI,CAAC,iBAAiB,CAAC,MAAM,CAAC,CAAC,CAAE,CAAC,IAAI,CAAC;YAAE,SAAQ;QACjD,8FAA8F;QAC9F,wBAAwB;QACxB,IAAI,oBAAoB,CAAC,MAAM,CAAC,CAAC,CAAE,CAAC,IAAI,CAAC;YAAE,SAAQ;QACnD,yFAAyF;QACzF,8BAA8B;QAC9B,IAAI,CAAC,GAAG,CAAC,CAAA;QACT,OACC,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC,MAAM;YACrB,iBAAiB,CAAC,MAAM,CAAC,CAAC,GAAG,CAAC,CAAE,CAAC,IAAI,CAAC;YACtC,CAAC,oBAAoB,CAAC,MAAM,CAAC,CAAC,GAAG,CAAC,CAAE,CAAC,IAAI,CAAC;YAC1C,CAAC,UAAU,CAAC,MAAM,CAAC,CAAC,GAAG,CAAC,CAAE,CAAC,IAAI,CAAC;YAChC,CAAC,cAAc,CAAC,MAAM,CAAC,CAAC,GAAG,CAAC,CAAE,CAAC,IAAI,CAAC,EACnC,CAAC;YACF,CAAC,EAAE,CAAA;QACJ,CAAC;QACD,uFAAuF;QACvF,uFAAuF;QACvF,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAA;QACrC,KAAK,IAAI,GAAG,GAAG,CAAC,EAAE,GAAG,IAAI,MAAM,EAAE,GAAG,EAAE,EAAE,CAAC;YACxC,MAAM,QAAQ,GAAG,MAAM,CAAC,CAAC,CAAE,CAAA;YAC3B,MAAM,MAAM,GAAG,MAAM,CAAC,CAAC,GAAG,GAAG,GAAG,CAAC,CAAE,CAAA;YACnC,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC,GAAG,CAAC,CAAA;YACvD,MAAM,YAAY,GAAG,GAAG,KAAK,CAAC,IAAI,eAAe,CAAC,GAAG,CAAC,QAAQ,CAAC,WAAW,EAAE,CAAC,CAAA;YAC7E,MAAM,MAAM,GAAG,CAAC,GAAG,GAAG,GAAG,CAAC,KAAK,MAAM,CAAC,MAAM,GAAG,CAAC,CAAA;YAChD,IAAI,UAAU,GAAG,IAAI,GAAG,CAAC,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAA;YACtE,IAAI,YAAY,IAAI,CAAC,MAAM;gBAAE,UAAU,IAAI,GAAG,CAAA;YAC9C,IAAI,MAAM,IAAI,aAAa;gBAAE,UAAU,IAAI,GAAG,CAAA;YAC9C,IAAI,MAAM;gBAAE,UAAU,IAAI,IAAI,CAAA;YAC9B,GAAG,CAAC,IAAI,CAAC;gBACR,IAAI,EAAE,WAAW,CAAC,IAAI,EAAE,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC,GAAG,CAAC;gBACnD,cAAc,EAAE,iBAAiB;gBACjC,UAAU,EAAE,IAAI,CAAC,GAAG,CAAC,IAAI,EAAE,UAAU,CAAC;aACtC,CAAC,CAAA;QACH,CAAC;QACD,wFAAwF;QACxF,0FAA0F;QAC1F,4FAA4F;IAC7F,CAAC;IACD,OAAO,GAAG,CAAA;AACX,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,gBAAgB,CAC/B,MAAmC,EACnC,IAAY,EACZ,cAAwB;IAExB,MAAM,GAAG,GAAqB,EAAE,CAAA;IAChC,IAAI,CAAC,GAAG,CAAC,CAAA;IACT,OAAO,CAAC,GAAG,MAAM,CAAC,MAAM,EAAE,CAAC;QAC1B,IAAI,CAAC,iBAAiB,CAAC,MAAM,CAAC,CAAC,CAAE,CAAC,IAAI,CAAC,EAAE,CAAC;YACzC,CAAC,EAAE,CAAA;YACH,SAAQ;QACT,CAAC;QACD,IAAI,CAAC,GAAG,CAAC,CAAA;QACT,OAAO,CAAC,GAAG,CAAC,GAAG,MAAM,CAAC,MAAM,IAAI,CAAC,iBAAiB,CAAC,MAAM,CAAC,CAAC,GAAG,CAAC,CAAE,CAAC,IAAI,CAAC,IAAI,MAAM,CAAC,CAAC,GAAG,CAAC,CAAE,CAAC,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC;YAC/G,CAAC,EAAE,CAAA;QACJ,CAAC;QACD,MAAM,GAAG,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,CAAA;QAClC,MAAM,YAAY,GAAG,iBAAiB,CAAC,GAAG,CAAC,CAAA;QAC3C,MAAM,iBAAiB,GAAG,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAA;QAEnE,IAAI,YAAY,GAAG,CAAC,IAAI,CAAC,iBAAiB,IAAI,GAAG,CAAC,MAAM,IAAI,CAAC,CAAC,EAAE,CAAC;YAChE,MAAM,QAAQ,GAAG,GAAG,CAAC,CAAC,CAAE,CAAA;YACxB,MAAM,MAAM,GAAG,GAAG,CAAC,GAAG,CAAC,MAAM,GAAG,CAAC,CAAE,CAAA;YACnC,MAAM,UAAU,GAAG,YAAY,GAAG,CAAC,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,IAAI,CAAA;YACzD,GAAG,CAAC,IAAI,CAAC;gBACR,IAAI,EAAE,WAAW,CAAC,IAAI,EAAE,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC,GAAG,CAAC;gBACnD,cAAc,EAAE,cAAc;gBAC9B,UAAU;aACV,CAAC,CAAA;QACH,CAAC;aAAM,IAAI,cAAc,IAAI,GAAG,CAAC,MAAM,IAAI,CAAC,EAAE,CAAC;YAC9C,MAAM,SAAS,GAAG,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,cAAc,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAA;YACzD,MAAM,aAAa,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,CAAE,CAAC,IAAI,CAAC,CAAA;YAC9C,MAAM,OAAO,GAAG,aAAa,CAAC,GAAG,CAAC,CAAA;YAClC,IAAI,CAAC,SAAS,IAAI,CAAC,aAAa,IAAI,CAAC,OAAO,EAAE,CAAC;gBAC9C,MAAM,QAAQ,GAAG,GAAG,CAAC,CAAC,CAAE,CAAA;gBACxB,MAAM,MAAM,GAAG,GAAG,CAAC,GAAG,CAAC,MAAM,GAAG,CAAC,CAAE,CAAA;gBACnC,GAAG,CAAC,IAAI,CAAC;oBACR,IAAI,EAAE,WAAW,CAAC,IAAI,EAAE,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAC,GAAG,CAAC;oBACnD,cAAc,EAAE,cAAc;oBAC9B,UAAU,EAAE,GAAG,CAAC,MAAM,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,GAAG;iBACxC,CAAC,CAAA;YACH,CAAC;QACF,CAAC;QAED,CAAC,GAAG,CAAC,GAAG,CAAC,CAAA;IACV,CAAC;IACD,OAAO,GAAG,CAAA;AACX,CAAC"}

package/out/types.d.ts

@@ -0,0 +1,56 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ */
+/** Re-exports of the canonical types from `@mailwoman/core/pipeline`. */
+export type { LocaleHint, PhraseGrouper, PhraseKind, PhraseProposal } from "@mailwoman/core/pipeline";
+/** Re-export of the canonical `Section` type from `@mailwoman/core/types`. `Section = Span`. */
+export type { Section } from "@mailwoman/core/types";
+/**
+ * Minimal `NormalizedInput` shape consumed by `groupPhrases`. Compatible with
+ * `@mailwoman/normalize`'s output.
+ */
+export interface NormalizedInputLite {
+    raw: string;
+    normalized: string;
+    appliedLocale?: string;
+}
+/**
+ * Minimal `QueryShape` shape consumed by `groupPhrases`. Compatible with `@mailwoman/query-shape`'s
+ * output.
+ */
+export interface QueryShapeLike {
+    knownFormats: ReadonlyArray<{
+        format: string;
+        span: {
+            start: number;
+            end: number;
+        };
+        confidence: number;
+    }>;
+    segments?: ReadonlyArray<{
+        body: string;
+        index: number;
+        span?: {
+            start: number;
+            end: number;
+        };
+    }>;
+    tokenClasses?: ReadonlyArray<{
+        span: {
+            start: number;
+            end: number;
+            body: string;
+        };
+        class: string;
+        length: number;
+    }>;
+    characterClass?: string;
+    totalLength?: number;
+}
+export interface GroupPhrasesOpts {
+    /** Reserved for future tunables (e.g. confidence floor, per-kind biasing). Currently unused. */
+    confidenceFloor?: number;
+}
+//# sourceMappingURL=types.d.ts.map

package/out/types.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../types.ts"],"names":[],"mappings":"AAAA;;;;GAIG;AAEH,yEAAyE;AACzE,YAAY,EAAE,UAAU,EAAE,aAAa,EAAE,UAAU,EAAE,cAAc,EAAE,MAAM,0BAA0B,CAAA;AAErG,gGAAgG;AAChG,YAAY,EAAE,OAAO,EAAE,MAAM,uBAAuB,CAAA;AAEpD;;;GAGG;AACH,MAAM,WAAW,mBAAmB;IACnC,GAAG,EAAE,MAAM,CAAA;IACX,UAAU,EAAE,MAAM,CAAA;IAClB,aAAa,CAAC,EAAE,MAAM,CAAA;CACtB;AAED;;;GAGG;AACH,MAAM,WAAW,cAAc;IAC9B,YAAY,EAAE,aAAa,CAAC;QAC3B,MAAM,EAAE,MAAM,CAAA;QACd,IAAI,EAAE;YAAE,KAAK,EAAE,MAAM,CAAC;YAAC,GAAG,EAAE,MAAM,CAAA;SAAE,CAAA;QACpC,UAAU,EAAE,MAAM,CAAA;KAClB,CAAC,CAAA;IACF,QAAQ,CAAC,EAAE,aAAa,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,IAAI,CAAC,EAAE;YAAE,KAAK,EAAE,MAAM,CAAC;YAAC,GAAG,EAAE,MAAM,CAAA;SAAE,CAAA;KAAE,CAAC,CAAA;IAChG,YAAY,CAAC,EAAE,aAAa,CAAC;QAC5B,IAAI,EAAE;YAAE,KAAK,EAAE,MAAM,CAAC;YAAC,GAAG,EAAE,MAAM,CAAC;YAAC,IAAI,EAAE,MAAM,CAAA;SAAE,CAAA;QAClD,KAAK,EAAE,MAAM,CAAA;QACb,MAAM,EAAE,MAAM,CAAA;KACd,CAAC,CAAA;IACF,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB,WAAW,CAAC,EAAE,MAAM,CAAA;CACpB;AAED,MAAM,WAAW,gBAAgB;IAChC,gGAAgG;IAChG,eAAe,CAAC,EAAE,MAAM,CAAA;CACxB"}

package/out/types.js

@@ -0,0 +1,7 @@
+/**
+ * @copyright Sister Software
+ * @license AGPL-3.0
+ * @author Teffen Ellis, et al.
+ */
+export {};
+//# sourceMappingURL=types.js.map

package/out/types.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../types.ts"],"names":[],"mappings":"AAAA;;;;GAIG"}

package/package.json

@@ -0,0 +1,28 @@
+{
+  "name": "@mailwoman/phrase-grouper",
+  "version": "4.0.0",
+  "description": "Stage 2.7 of the runtime pipeline — propose coherent input units (boundary discovery) with a structural kind hypothesis + confidence. Rule-based v1 (port of v1 section/sub-section logic); learned 1-2M-param span proposer reserved for v0.5.1.",
+  "license": "AGPL-3.0-only",
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/sister-software/mailwoman.git",
+    "directory": "phrase-grouper"
+  },
+  "type": "module",
+  "exports": {
+    "./package.json": "./package.json",
+    ".": "./out/index.js"
+  },
+  "dependencies": {
+    "@mailwoman/core": "4.0.0"
+  },
+  "files": [
+    "out/**/*.js",
+    "out/**/*.js.map",
+    "out/**/*.d.ts",
+    "out/**/*.d.ts.map"
+  ],
+  "publishConfig": {
+    "access": "public"
+  }
+}