@mailwoman/neural-weights-it-it 9.4.0 → 10.1.0

package/LICENSE.md

@@ -0,0 +1,36 @@
+# License — @mailwoman/neural-weights-it-it
+
+Generated by `mwops release write-rights-files` from package.json and model-card.json. Edit the generator or those files, never this one.
+
+## The terms Sister Software grants
+
+This package is distributed under `AGPL-3.0-only OR LicenseRef-Commercial`.
+
+Under the AGPL-3.0-only branch, three obligations attach: keep the copyright notices, license a
+modified copy under the same terms, and offer corresponding source to people who use a modified copy
+over a network (section 13). Under a commercial agreement, the source offer and the share-alike
+condition do not apply to the code and model artifacts Sister Software authors.
+
+- Open-source terms: https://github.com/sister-software/mailwoman/blob/main/LICENSE.md
+- Commercial terms, as executed at checkout: https://mailwoman.ai/license/terms/commercial-2026-10
+- Commercial terms, as a reference template that does not itself execute:
+  https://github.com/sister-software/mailwoman/blob/main/COMMERCIAL-LICENSE.md
+
+## What the commercial grant does not cover
+
+Section 2 of the commercial agreement limits the grant to the code and model artifacts Sister
+Software authors. It does not extend to third-party components distributed with Mailwoman, which
+stay under their own licenses. Any attribution or share-alike obligation attaching to those
+components continues to apply and is not waived or relicensed by a commercial agreement.
+
+## Where the sources are recorded
+
+`PROVENANCE.json` in this package records, per artifact, what the repository holds about its
+inputs and which questions it cannot answer. Read it as the record it is: a package with no recorded
+training attribution is a package whose model card records none, which is not the same as a package
+with no attributable inputs. No artifact in this package is recorded as cleared for any particular
+use.
+
+`THIRD_PARTY_NOTICES.md` in the repository lists the components Mailwoman distributes and the terms
+each one carries. Reference data an operator downloads separately carries the terms of whoever
+published it, and nothing here states what those are.

package/PROVENANCE.json

@@ -0,0 +1,119 @@
+{
+	"$comment": "Generated by `mwops release write-rights-files` from package.json and model-card.json. Edit the generator or those files, never this one.",
+	"package": "@mailwoman/neural-weights-it-it",
+	"package_version": "10.0.0",
+	"license": "AGPL-3.0-only OR LicenseRef-Commercial",
+	"model_card_version": "7.0.0",
+	"version_series": "overlay",
+	"version_series_meaning": "This package is a data-only overlay, and model_card_version names this overlay's own artifacts. The model graph its rows decode through is the one base_weights names, at that package's own model_card_version. Neither number is package_version, which moves in lockstep across every mailwoman workspace on every release.",
+	"base_weights": "@mailwoman/neural-weights-en-us",
+	"inherited_lineage": {
+		"package": "@mailwoman/neural-weights-en-us",
+		"package_version": "10.0.0",
+		"model_card_version": "9.1.0",
+		"chain": ["@mailwoman/neural-weights-it-it", "@mailwoman/neural-weights-en-us"],
+		"note": "These entries are @mailwoman/neural-weights-en-us's record of what its model graph was trained on. This package ships no graph and contributed none of these rows. Read them as the base's lineage rather than as a claim about this package's own artifacts, which are listed under artifacts and attributed under training_attribution.",
+		"attribution": [
+			{
+				"text": "HM Land Registry — Price Paid Data (OGL v3.0): the `synth-gb-v1` training extract (800,000 rows) NEW in 6.7.0. Crown copyright, attribution required. The GB placetype-pair index is built from the same source and ships in `neural-weights-en-gb`, whose own card carries its attribution.",
+				"license_named": "OGL v3.0"
+			},
+			{
+				"text": "LINZ-derived OpenAddresses NZ (CC-BY 4.0): the `synth-nz-v2` training extract (800,000 rows) NEW in 6.7.0.",
+				"license_named": "CC-BY 4.0"
+			},
+			{
+				"text": "OpenAddresses ES tuples (CC-BY per OA source): the `synth-es-pedania-v1` training extract (800,000 rows, below-municipio pedanía place names) NEW in 6.7.0.",
+				"license_named": "CC-BY per OA source"
+			},
+			{
+				"text": "BAN — Base Adresse Nationale (DINUM / IGN, Licence Ouverte 2.0): the `synth-fr-lieudit-v1` training extract (`nom_ld` lieu-dit field) NEW in 6.7.0, additional to the existing `synth-fr-fragment` extract (144,865 rows, 6.4.0).",
+				"license_named": "DINUM / IGN, Licence Ouverte 2.0"
+			},
+			{
+				"text": "Overture Maps Foundation — Addresses theme (per-country register license; Overture declares no identifier of its own for this theme, so the terms are the upstream register's): the `overture-latam` CA/MX training extract (456,230 rows, 6.6.0). Attribution-only, not share-alike. Read 2026-09-25 from https://docs.overturemaps.org/attribution/.",
+				"license_named": "456,230 rows, 6.6.0"
+			},
+			{
+				"text": "OpenAddresses CZ — ČÚZK RÚIAN (CC-BY 4.0): tokenizer-splice training text + the oa-cz coord eval sets",
+				"license_named": "CC-BY 4.0"
+			},
+			{
+				"text": "OpenAddresses PL — GUGiK / PRG (public, BDOT-derived): tokenizer-splice training text + the oa-pl coord eval sets",
+				"license_named": null
+			},
+			{
+				"text": "OpenAddresses SK — MINV/ZBGIS (CC-0/CC-BY per OA source): oa-sk coord eval set (free-rider validation)",
+				"license_named": "CC-0/CC-BY per OA source"
+			},
+			{
+				"text": "OpenAddresses SI — GURS (CC-BY per OA source): oa-si coord eval set (free-rider validation)",
+				"license_named": "CC-BY per OA source"
+			},
+			{
+				"text": "See THIRD_PARTY_NOTICES.md for the standing OpenAddresses / GeoNames / WOF attribution.",
+				"license_named": null
+			}
+		],
+		"unresolved": null
+	},
+	"artifacts": [
+		{
+			"path": "pair-index-it.bin",
+			"role": "placetype-pair-index",
+			"md5": null,
+			"digest": "unrecorded"
+		},
+		{
+			"path": "fst-it-it.bin",
+			"role": "gazetteer-fst",
+			"md5": null,
+			"digest": "unrecorded"
+		},
+		{
+			"path": "fst-street-morphology.bin",
+			"role": "gazetteer-fst",
+			"md5": null,
+			"digest": "unrecorded"
+		},
+		{
+			"path": "anchor-lexicon-v1.json",
+			"role": "lexicon",
+			"md5": null,
+			"digest": "unrecorded"
+		},
+		{
+			"path": "locality-surface-lexicon-v7.json",
+			"role": "lexicon",
+			"md5": null,
+			"digest": "unrecorded"
+		},
+		{
+			"path": "street-type-lexicon-v3.json",
+			"role": "lexicon",
+			"md5": null,
+			"digest": "unrecorded"
+		},
+		{
+			"path": "country-surface-lexicon-v1.json",
+			"role": "lexicon",
+			"md5": null,
+			"digest": "unrecorded"
+		}
+	],
+	"model_graph": null,
+	"training_attribution": {
+		"status": "none-recorded-in-this-package",
+		"entries": [],
+		"by_use": {}
+	},
+	"attribution_recorded_in_another_package": [],
+	"corpus_version": null,
+	"effective_training_sources": null,
+	"tokenizer_version": null,
+	"unresolved": [
+		"No digest is recorded for 7 of 7 declared artifacts: pair-index-it.bin, fst-it-it.bin, fst-street-morphology.bin, anchor-lexicon-v1.json, locality-surface-lexicon-v7.json, street-type-lexicon-v3.json, country-surface-lexicon-v1.json. An unrecorded digest is not a verified one.",
+		"This package's model card records no training attribution of its own. It ships 7 artifacts and no model graph, so its own inputs are unattributed here while the 10 entries under inherited_lineage describe @mailwoman/neural-weights-en-us's graph.",
+		"Which records trained the model this package ships or inherits is unresolved. `buildCorpus` writes a frozen `TRAINING_SOURCES.json` naming every source that contributed and under which terms, `effective-manifest.run.ts` derives from it and an `audit_epoch_mixture` output what one config's epoch drew, and no released model was built from a corpus carrying either. The corpus named above is the recipe's corpus rather than a record of what reached the model."
+	]
+}

package/README.md

@@ -8,7 +8,7 @@ it-IT weights overlay for [mailwoman](https://mailwoman.ai).
 It also ships the four evidence lexicons the base model was trained against — `anchor-lexicon-v1.json`,
 `country-surface-lexicon-v1.json`, `street-type-lexicon-v3.json`, `locality-surface-lexicon-v7.json` — and
 its `model-card.json` carries the base card's `requires` block, so a consumer runs the same channels the
-board grades. A data-only overlay without them ran gazetteer, country and both evidence channels OFF
+board grades. A data-only overlay without them ran gazetteer, country and both evidence channels off
 against a model trained with them (#2115).
 
 ```sh

package/model-card.json

@@ -4,7 +4,7 @@
 	"locale": "it-IT",
 	"license": "AGPL-3.0",
 	"requires": {
-		"$comment": "Copy of the base card's requires block — same trained ship-config, same fail-closed contract via createScorer. This overlay shares the base model.onnx byte-for-byte, so the channels it was trained against are the base's; keep in lockstep with neural-weights-en-us/model-card.json on every base ship. Without this block the loader inferred the channels from the ONNX graph and resolved the evidence lexicons by legacy filename, which fed street_type by guess and left locality_surface OFF (v6 sought, v7 shipped); a consumer of the published tarball, which carried no lexicon at all, ran gazetteer, country and both evidence channels OFF against a model trained with them (#1668, #2115). Measured before shipping: locality_surface ON vs OFF, 0 of 49 board rows differ (DE 12, ES 25, IT 12).",
+		"$comment": "Copy of the base card's requires block — same trained ship-config, same fail-closed interface via createScorer. This overlay shares the base model.onnx byte-for-byte, so the channels it was trained against are the base's; keep in lockstep with neural-weights-en-us/model-card.json on every base ship. Without this block the loader inferred the channels from the ONNX graph and resolved the evidence lexicons by legacy filename, which fed street_type by guess and left locality_surface OFF (v6 sought, v7 shipped); a consumer of the published tarball, which carried no lexicon at all, ran gazetteer, country and both evidence channels OFF against a model trained with them (#1668, #2115). Measured before shipping: locality_surface ON vs OFF, 0 of 49 board rows differ (DE 12, ES 25, IT 12).",
 		"anchor": {
 			"required": true,
 			"span_mode": "shaped",
@@ -51,7 +51,7 @@
 			"bytes": 107845,
 			"$comment_pix2": "PIX2 (schema 3, 2026-08-04): every record now carries the PARENT's ComponentTag alongside the child's — +1 byte per pair, pair COUNT unchanged. The recorded parent tag is the WOF parent ROW's placetype projected through PLACETYPE_PROJECTION, so a locality/localadmin parent reads `locality` and a borough parent reads `dependent_locality`. This artifact deliberately ships WITHOUT a `parentDelta`: no board has graded the parent side for this locale, and the D-rule's answer to an unmeasured locale is a per-locale check rather than a magnitude inherited from us/gb/nz/fr. Absent means the parent bias is OFF here, not zero.",
 			"delta_calibration": "δ=10, transitionBeta=5 — inherited from GB per R5's uniform-deficit finding.",
-			"country_check": "hard-blocked on the header's country (it).",
+			"country_check": "hard-enforced: applies only when the header's country matches (it).",
 			"requires_leading_postcode_strip": "INERT without the `it` entries in SEGMENT_PARENT_POSTCODE_SHAPES and LEADING_POSTCODE_COUNTRIES (neural/placetype-pair-prior.ts) — Italian addresses write the postcode first.",
 			"parent_alias_expansion": "ENABLED. WOF stores `Rome` (its English preferred name) for a city Italians write `Roma`, carried as the `ita` preferred name. Aliases are drawn from names whose language is OFFICIAL for the country (codex isOfficialLanguage) plus English, and must be LATIN-SCRIPT — the model never sees other scripts.",
 			"not_committed": "derived artifact, .gitignore'd — rebuilt via `mailwoman gazetteer pair-index --out neural-weights-it-it --country it --delta 10 --transition-beta 5 --borough-db <wof/admin-global-priority.db>`"

package/package.json

@@ -1,8 +1,8 @@
 {
   "name": "@mailwoman/neural-weights-it-it",
-  "version": "9.4.0",
+  "version": "10.1.0",
   "description": "it-IT weights overlay for mailwoman — data-only; shares the base model with @mailwoman/neural-weights-en-us.",
-  "license": "AGPL-3.0",
+  "license": "AGPL-3.0-only OR LicenseRef-Commercial",
   "repository": {
     "type": "git",
     "url": "https://github.com/sister-software/mailwoman.git",
@@ -10,7 +10,11 @@
   },
   "files": [
     "model-card.json",
+    "LICENSE.md",
+    "PROVENANCE.json",
     "pair-index-it.bin",
+    "fst-it-it.bin",
+    "fst-street-morphology.bin",
     "anchor-lexicon-v1.json",
     "locality-surface-lexicon-v7.json",
     "street-type-lexicon-v3.json",
@@ -24,21 +28,14 @@
     "!test/**"
   ],
   "type": "module",
-  "imports": {
-    "#*": {
-      "types": "./out/*.d.ts",
-      "node": "./out/*.js",
-      "default": "./out/*.js"
-    }
-  },
   "exports": {
     "./*": "./*"
   },
   "dependencies": {
-    "@mailwoman/neural-weights-en-us": "9.4.0"
+    "@mailwoman/neural-weights-en-us": "10.1.0"
   },
   "devDependencies": {
-    "@mailwoman/resolver-wof-sqlite": "9.4.0"
+    "@mailwoman/resolver-wof-sqlite": "10.1.0"
   },
   "mailwoman": {
     "baseWeights": "@mailwoman/neural-weights-en-us"
@@ -46,13 +43,6 @@
   "publishConfig": {
     "exports": {
       "./*": "./*"
-    },
-    "imports": {
-      "#*": {
-        "types": "./out/*.d.ts",
-        "node": "./out/*.js",
-        "default": "./out/*.js"
-      }
     }
   }
 }