@mailwoman/corpus 10.0.0 → 10.1.0

package/data/licenses/PROVENANCE.md

@@ -0,0 +1,65 @@
+# `corpus/data/licenses/` — retrieved license texts
+
+This directory holds the copies that license decisions are read from. `ElectedLicense.retrievedCopy` in the source
+register points to a file here, so a later reviewer checks the same text instead of the page as it reads that day.
+
+A publisher's page changes over time. A license is versioned but the page that carries it is not, so a decision
+recorded against the page cannot be rechecked once the page changes. Each file here holds the text as retrieved, with
+the URL and the retrieval date at its head, and is never edited after that. A correction requires a new retrieval.
+
+These files elect no license. They are the texts a license reading would use, and every license decision in
+`address-source-register.json` still reads `unchecked`.
+
+## What is here
+
+| file                           | publisher                                      | retrieved  | how                                     |
+| ------------------------------ | ---------------------------------------------- | ---------- | --------------------------------------- |
+| `licence-ouverte-2.0.md`       | Etalab, for the French government              | 2026-09-21 | `etalab/licence-ouverte`, `LO.md`       |
+| `sirene-publication.md`        | INSEE, via data.gouv.fr                        | 2026-09-21 | the dataset page                        |
+| `whosonfirst-licenses.md`      | Who's On First, originally Mapzen              | 2026-09-21 | the docs page and the repo `LICENSE.md` |
+| `geonames-publication.md`      | GeoNames                                       | 2026-09-21 | the export readme and the about page    |
+| `cdla-permissive-2.0.md`       | The Linux Foundation, stewarding the agreement | 2026-09-21 | the SPDX license list, plain text       |
+| `nlod-2.0.md`                  | the responsible Norwegian ministry             | 2026-09-21 | the SPDX license list, plain text       |
+| `national-address-database.md` | United States Department of Transportation     | 2026-09-21 | the database's own disclaimer page      |
+
+The first filename keeps the publisher's own spelling, because it is the name of the grant.
+
+Five of the seven files are here although no register source points to them. The four runtime bundles' rights records
+cite those terms, and an operator acts on them today. A file that covers a bundle instead of a register source is the
+ordinary case. The terms for a source that nobody can ingest can wait, but the terms for a database that somebody
+downloads this week cannot.
+
+The National Address Database is the strongest reason for putting bundles first. Its rows are 85,482,142 of the `us` bundle's
+125,276,536, which is 68.2%. The bundle's record mentioned neither the database nor the Department of Transportation
+until a census of the shipped databases' own `source` column found them.
+
+Three of the five retrievals changed a record instead of only supporting one. The Who's On First page lists Ordnance
+Survey of Northern Ireland and Ordnance Survey Ireland and never mentions Royal Mail, so half of the `candidate`
+bundle's open question concerned a body absent from the text. CDLA-Permissive-2.0 §3.1 states that the agreement
+imposes no restriction or obligation on Results, and §5.4 defines Results to include machine learning models. The
+`poi` bundle's conditions therefore separate what redistributing the database requires from what the model carries.
+The National Address Database's page states both that the data carries no copyright under 17 U.S.C. § 105 and that it
+is not intended for use as a mailing list under state statutes. A record that kept only the first statement would lose
+the second.
+
+A quotation of a grant written in British English sits in a fenced block instead of a blockquote, so that the
+publisher's own spelling survives a prose sweep. `nlod-2.0.md` is an example.
+
+## What could not be retrieved, and what that means
+
+A failed automated fetch leaves a text unretrieved, and it does not indicate a prohibition. The register records
+that failure separately from a license restriction. A register that listed only decisions would show a source with
+no archived terms and a source whose terms refuse use in the same way.
+
+| target                   | attempt                                                    | result                         |
+| ------------------------ | ---------------------------------------------------------- | ------------------------------ |
+| AusTender copyright page | `www.tenders.gov.au/Content/Copyright`, retried 2026-09-21 | HTTP 403 to an automated fetch |
+
+That page needs a person with a browser or an authenticated retrieval. The result makes no statement about what those
+terms grant.
+
+**One earlier entry is withdrawn.** NLOD 2.0 was recorded here as returning HTTP 406 from
+`data.norge.no/nlod/en/2.0`. The same URL returned HTTP 200 on 2026-09-21, and the SPDX license list
+serves the same text as plain text, so the text is retrieved and archived as `nlod-2.0.md`. A single
+failed fetch describes one attempt, and the earlier entry was wrong to record it as a property of the
+endpoint. Retry before recording a failure in a durable record.

package/data/licenses/cdla-permissive-2.0.md

@@ -0,0 +1,43 @@
+# Community Data License Agreement — Permissive, Version 2.0 — as retrieved
+
+- Source: `https://raw.githubusercontent.com/spdx/license-list-data/main/text/CDLA-Permissive-2.0.txt`
+- Retrieved: 2026-09-21
+- Publisher: The Linux Foundation, as the steward of the agreement
+- Version stated in the text: 2.0
+
+This is the grant the Overture Maps Foundation places its Places theme under, which is the `poi`
+runtime bundle. The text was taken from the SPDX license list rather than from `cdla.dev`, because
+SPDX publishes it as plain text and the two carry the same words.
+
+## Data-sharing terms (quoted)
+
+> 2.1. A Data Recipient may share Data, with or without modifications, so long as the Data Recipient
+> makes available the text of this agreement with the shared Data.
+
+That is the one condition the agreement places on redistributing the database, and it is a condition
+to supply this text rather than to name a contributor. This repository hosts `poi.db` for download,
+so the condition attaches to that hosting, which is why this file is committed.
+
+## A model trained on the data, quoted
+
+> 3. No Restrictions on Results
+>
+> 3.1. This agreement does not impose any restriction or obligations with respect to the use,
+> modification, or sharing of Results.
+
+> 5.4. "Results" means any outcome obtained by computational analysis of Data, including for example
+> machine learning models and models' insights.
+
+Read together, the agreement states that a machine-learning model is a Result and that it carries no
+restriction or obligation under the agreement. Among the grants this repository reads, this is the
+one that addresses a trained model in its own text rather than leaving it to be argued from the
+definition of a derived work.
+
+Two things that does not establish. It makes no statement about the terms of whatever sources Overture
+itself drew each row from, which the `poi` bundle's record still carries as an open question. And it
+is a statement about this agreement alone, so it does not decide the question for any other source.
+
+## What this text does not settle
+
+No election is made here. The `poi` bundle's record cites these terms and no license decision in
+`address-source-register.json` reads anything but `unchecked`.

package/data/licenses/geonames-publication.md

@@ -0,0 +1,34 @@
+# GeoNames publication terms — as retrieved
+
+- Source: `https://download.geonames.org/export/dump/readme.txt`, with `https://www.geonames.org/about.html`
+  retrieved on the same date
+- Retrieved: 2026-09-21
+- Publisher: GeoNames
+- Version stated in the text: Creative Commons Attribution 4.0, in the export readme
+
+GeoNames folds into the `candidate` runtime bundle, whose record names its terms. Two of the
+publisher's own pages state them and only one gives a version.
+
+## What the export readme states, quoted in full
+
+> This work is licensed under a Creative Commons Attribution 4.0 License,
+> see https://creativecommons.org/licenses/by/4.0/
+> The Data is provided "as is" without warranty or any representation of accuracy, timeliness or
+> completeness.
+
+## What the about page states
+
+> The GeoNames geographical database is available for download free of charge under a creative
+> commons attribution license.
+
+The about page names no version. The readme beside the files themselves names 4.0. That readme travels
+with the data, so the bundle's record states 4.0 and names this file.
+
+Both sources matter because the two pages give different information. The about page alone leaves the
+version unknown. Creative Commons Attribution's conditions differ across versions.
+
+## What this text does not settle
+
+Which GeoNames records reached the `candidate` bundle, and therefore what has to be attributed for
+which rows. GeoNames itself integrates data from other sources, which its about page states in those
+words, and this text makes no statement about the terms of those. No election is made here.

package/data/licenses/licence-ouverte-2.0.md

@@ -0,0 +1,36 @@
+# Licence Ouverte 2.0 — as retrieved
+
+- Source: `https://raw.githubusercontent.com/etalab/licence-ouverte/master/LO.md`
+- Retrieved: 2026-09-21
+- Publisher: Etalab, for the French government
+- Version stated in the text: 2.0
+
+The canonical page `https://www.etalab.gouv.fr/licence-ouverte-open-licence/` now redirects to
+`https://www.data.gouv.fr/`, so the text was taken from Etalab's own repository, which is where the page sourced it.
+
+## Rights granted, quoted
+
+> droit non exclusif et gratuit de libre « Réutilisation » de l'« Information »
+
+The grant covers commercial and non-commercial purposes. It permits communicating, reproducing,
+adapting, modifying, distributing, and exploiting the information commercially.
+
+## The condition, quoted
+
+> mentionner la paternité de l'«Information» : sa source (a minima le nom du « Concédant ») et la date de la dernière
+> mise à jour
+
+## Conditions and exclusions the text states
+
+- Attribution must not suggest the licensor endorses the reuse.
+- Reuse of personal data must comply with the data-protection frameworks that apply to it.
+- The reuser must not mislead a third party about the content, source or update date of the information.
+- The information is provided `telle que produite ou reçue, sans autre garantie expresse ou tacite`, with no warranty
+  against defects, errors, or continued availability.
+- The reuser carries sole responsibility for the reuse.
+
+## What this text does not settle
+
+It grants reuse of the information. Whether model weights trained on that information are themselves a reuse under this
+grant, and what attribution discharges the condition for a trained model where no individual record is recoverable, are
+readings the text does not make. The grant carries no share-alike condition.

package/data/licenses/national-address-database.md

@@ -0,0 +1,46 @@
+# National Address Database (NAD) disclaimer — as retrieved
+
+- Source: `https://www.transportation.gov/gis/national-address-database/national-address-database-nad-disclaimer`
+- Retrieved: 2026-09-21
+- Publisher: the United States Department of Transportation
+- Version stated in the text: none. The page states it was last updated 2026-01-09.
+
+The National Address Database supplies 85,482,142 of the 125,276,536 address points in the `us`
+runtime bundle, which is 68.2% of it. The bundle's record mentioned neither the database nor its
+publisher until this text was read. The rows reach the bundle through the Overture Maps Foundation.
+Overture stamps each row `NAD`, so the bundle carries the publisher on every row.
+
+## What the disclaimer states about reuse, quoted
+
+> The NAD is a work of the federal government and as such is not subject to copyright protection in
+> accordance with 17 U.S.C. § 105. It is available for re-use without limitation or restriction.
+
+## The restriction that is not a copyright restriction, quoted
+
+> Although the NAD is open data, it is not intended for use as a mailing list and is subject to
+> individual state statutes that prohibit it's use as a mailing list.
+
+These two sentences appear on the same page, and a record that says only "no copyright under 17
+U.S.C. § 105" would lose the second one. The first sentence says that copyright imposes no
+restriction. The second states a restriction from another source: individual state statutes, which
+the page neither names nor lists, on one particular use. A postal address product is close to that
+use, so the restriction belongs in the record and not in a footnote.
+
+This text does not establish which states have those statutes, what they prohibit, or whether any of
+them applies to what this repository does with the rows.
+
+## What the disclaimer states about NAD's own sources, quoted
+
+> The National Address Database (NAD) is, with few exceptions, aggregated from state data, which is
+> in turn aggregated from local data.
+
+> The states that partner with USDOT to provide data for the NAD do so willingly, with no obligation.
+> Not every state that provides data to the NAD has complete coverage.
+
+NAD is therefore itself an aggregate. The federal-work statement covers NAD as published and does not
+cover each contributing state's own publication.
+
+## What this text does not settle
+
+The text does not say which state or local body supplied a given NAD row, or what that body requires.
+The bundle's rows carry `overture:NAD` and record no finer source. This file elects no license.

package/data/licenses/nlod-2.0.md

@@ -0,0 +1,90 @@
+# `Norwegian Licence for Open Government Data (NLOD) 2.0` — as retrieved
+
+- Source: `https://raw.githubusercontent.com/spdx/license-list-data/main/text/NLOD-2.0.txt`, with
+  `https://data.norge.no/nlod/en/2.0` returning HTTP 200 on the same date
+- Retrieved: 2026-09-21
+- Publisher: the responsible Norwegian ministry, named in the text as the Ministry of Local
+  Government and Modernisation
+- Version stated in the text: 2.0
+
+An earlier pass recorded `data.norge.no/nlod/en/2.0` as returning HTTP 406 to an automated fetch. It
+returned HTTP 200 on 2026-09-21, so that record was a transient failure rather than a refusal, and
+`PROVENANCE.md` is corrected. The text below is from SPDX, which publishes it as plain text.
+
+This is the grant the register's Brønnøysund entity register row would be read against.
+
+Every quotation below is in a fenced block so the publisher's own spelling and wording survive a
+prose sweep. No text inside a fence is edited.
+
+## What the license grants
+
+```text
+2. Licence
+
+The licensee, subject to the limitations that follow from this licence, may use the information for
+any purpose and in all contexts, by:
+
+     * copying the information and distributing the information to others,
+     * modifying the information and/or combining the information with other information, and
+     * copying and distributing such changed or combined information.
+
+This is a non-exclusive, free, perpetual and worldwide licence. The information may be used in any
+medium and format known today and/or which will become known in the future. The Licensee shall not
+sub-license or transfer this licence.
+```
+
+"For any purpose and in all contexts" reaches commercial use. The last sentence is the one to read
+before any commercial-sublicense reading: the licensee may not sub-license the grant itself.
+
+## The exemptions
+
+```text
+3. Exemptions
+The licence does not apply to and therefore does not grant a right to use:
+
+     * information which contains personal data covered by the Norwegian Personal Data Act unless
+       there is a legitimate basis for the disclosure and further processing of the personal data
+     * information distributed in violation of a statutory obligation to observe confidentiality
+     * information excluded from public disclosure pursuant to law, including information deemed
+       sensitive under the Norwegian National Security Act
+     * information subject to third party rights which the licensor is not authorised to license to
+       the licensee
+     * information protected by intellectual property rights other than copyright and neighbouring
+       rights in accordance with Chapter 5 of the Norwegian Copyright Act, such as trademarks,
+       patents and design rights …
+
+If the licensor has made available information not covered by the licence according to the above
+list, the licensee must cease all use of the information under the licence, and erase the
+information as soon as he or she becomes aware of or should have understood that the information is
+not covered by the licence.
+```
+
+The first exemption carves personal data out of the grant itself. A source under this license is
+therefore not fully granted by the license alone, which is why `AddressSourceRecord` carries a
+`personalDataReview` apart from its license decision.
+
+## The attribution condition
+
+```text
+5. Attribution
+The licensee shall attribute the licensor as specified by the licensor and include a reference to
+this licence. To the extent practically possible, the licensee shall provide a link to both this
+licence and the source of the information.
+
+If the licensor has not specified how attributions shall be made, the licensee shall normally state
+the following: «Contains data under the Norwegian licence for Open Government data (NLOD)
+distributed by [name of licensor]».
+
+If the information has been changed, the licensee must clearly indicate that changes have been made
+by the licensee.
+```
+
+Four requirements in one section: attribute as the licensor specifies, reference the license, link to
+both the license and the source where practical, and state that the information was changed. A record
+carrying only "attribution required" loses the last three.
+
+## What this text does not settle
+
+Whether a statistical model trained on information under this license is itself covered. The text
+does not address a model, where CDLA-Permissive-2.0 does. No election is made here, and every license
+decision in `address-source-register.json` still reads `unchecked`.

package/data/licenses/sirene-publication.md

@@ -0,0 +1,38 @@
+# SIRENE — the publication, as retrieved
+
+- Source: `https://www.data.gouv.fr/datasets/base-sirene-des-entreprises-et-de-leurs-etablissements-siren-siret/`
+- Retrieved: 2026-09-21
+- Publisher: Institut national de la statistique et des études économiques (INSEE)
+- Terms named on the page: `Licence Ouverte / Open Licence version 2.0`, archived beside this as
+  `licence-ouverte-2.0.md`
+- Last update stated on the page: 2026-09-01
+
+## What the publication offers
+
+Twelve primary files, as six pairs in ZIP and Parquet: the stock of legal units, active and ceased; historicized values
+of legal units; the stock of establishments, active and closed; historicized values of establishments; establishment
+succession links; and SIREN duplicates. Twelve documentation files accompany them, in PDF and CSV.
+
+## Personal data, quoted and paraphrased
+
+The page states the dataset carries personal data subject to the GDPR and French data-protection law, and that a reuser
+must respect its dissemination-status codes.
+
+For a sole trader or a legal representative who has filed an opposition, INSEE masks specific fields:
+
+> the identity of the entrepreneur (name, first names…), the address in the municipality and geolocation will be masked
+
+Legal-representative contact information is not published by INSEE at all, whatever the opposition status, under
+article R 123-232 of the commercial code.
+
+## What this page does not settle
+
+**Territory coverage.** The register maps nine overseas jurisdictions to SIRENE — BL, GF, GP, MF, MQ, PM, RE, WF and
+YT — and this page lists national stock files rather than a per-territory breakdown. Whether every one of those nine is
+present in the resource is a question the files answer and this page does not. Counsel's direction is explicit that the
+mapping is not evidence of presence.
+
+**Whether the masking suffices for what we do.** INSEE masks an opposed sole trader's identity, municipality address
+and geolocation at the source, which is a stronger starting position than a dataset with no opt-out. Whether a model
+trained on the unmasked remainder can memorize an address, and what measures that calls for, is a separate reading from
+the license.

package/data/licenses/whosonfirst-licenses.md

@@ -0,0 +1,65 @@
+# Who's On First data licenses — as retrieved
+
+- Source: `https://whosonfirst.org/docs/licenses/`, and the dataset repository's own
+  `https://raw.githubusercontent.com/whosonfirst/whosonfirst-data/master/LICENSE.md`
+- Retrieved: 2026-09-21
+- Publisher: Who's On First, originally Mapzen
+- Version stated in the text: none. The page states it was generated by `bin/docs.py` on 2020-02-21.
+
+Two pages carry the terms. They differ in length but agree in substance. The repository's
+`LICENSE.md` is four paragraphs and links to the page. The page repeats those paragraphs and adds the
+list of contributing sources.
+
+## What the repository's `LICENSE.md` states, quoted in full
+
+> Crediting Who's On First is recommended and linking back to the License is required. For example:
+>
+> > Data from Who's On First. [License](https://whosonfirst.org/docs/licenses/).
+>
+> The Who's On First dataset is both an original work and a modification of existing open data. Some
+> of those open data projects do require attribution. We have listed all sources in the full
+> [license file](https://whosonfirst.org/docs/licenses/).
+>
+> Remember, some sources require attribution, some do not. The format and structure that allows Who's
+> On First to operate is made available under a
+> [Creative Commons Zero](https://creativecommons.org/choose/zero/) designation, and
+> [a shout out](https://twitter.com/mapzen/) would be lovely.
+
+The wording separates three terms that are easy to confuse.
+
+The project **recommends** crediting Who's On First and **requires** a link to the license. The
+Creative Commons Zero designation covers the **format and structure**, in those words, and does not
+cover the records.
+
+## The contributing sources, measured
+
+The page lists **102** sources, each with a link to its data and a link to its license. The count was
+taken on 2026-09-21 by counting the `<name>: source data and license` entries on the retrieved page.
+
+The page states that its list was generated on 2020-02-21. The dataset has changed since then, so the
+list shows what contributed up to that date and does not show what a current bundle contains.
+
+Four entries bear on questions this repository has open:
+
+| entry                               | the license the page links      |
+| ----------------------------------- | ------------------------------- |
+| Ordnance Survey of Northern Ireland | `Open Government Licence v3.0`  |
+| Ordnance Survey Ireland             | the publisher's own portal page |
+| Yahoo! GeoPlanet                    | the publisher's own page        |
+| GeoNames                            | the publisher's own page        |
+
+**Royal Mail does not appear anywhere on the page.** The `candidate` bundle's record carried an open
+question about "the Ordnance Survey and Royal Mail notice". The page does list the two Ordnance
+Survey bodies, but it does not mention Royal Mail.
+
+A separate census already addressed the GeoPlanet entry. Across the four Who's On First
+bundles in the local data root, **zero of 7,189,494 `name:*` variant keys carry the `_p` GeoPlanet
+suffix**, and neither postalcode bundle carries a `name:*` key at all, so no GeoPlanet-derived string
+reaches a corpus row through those bundles.
+
+## What this text does not settle
+
+The text does not say which of the 102 sources contributed to a given record. The page lists the
+sources for the dataset as a whole, and the per-record `src:geom` property records a geometry source
+and not a name source. This file elects none of these terms, and every license decision in `address-source-register.json`
+still reads `unchecked`.