notes: decisions/0005-2026-11-03-what-kskb-shows-of-a-source

This data as json

path title type status classification words body
decisions/0005-2026-11-03-what-kskb-shows-of-a-source ADR 0005: what KSKB shows of a source follows the source's license adr proposed 2026-11-03 on Mark's word; adopted when the pull request that carries it is merged; the shared text database goes live at kskb.dev after that merge; the KERI Foundation's own content stays at abstract until Henk van Cann names the Creative Commons version partner 1591 # ADR 0005: what KSKB shows of a source follows the source's license > **ADR 0005.** *Status:* proposed 2026-11-03. *Decision:* what KSKB (the KERISuite Knowledge Base) shows of a source, on kskb.dev and in its shared copies, is set by that source's own license, read and recorded source by source, and by the kind of document. Where the license allows reproduction and the document is prose, the text is shown in full beside KSKB's abstract, with the credit and the notices the license asks for. Where no license is found, KSKB shows metadata, its own abstract, a link and at most a short marked quotation. Conversation (the Slack archive, repository issues and pull-request threads) is indexed and never republished. The local corpus stays a gitignored working copy. This replaces the blanket rule of ADR 0002 that the text is never shared, and decision 3 of ADR 0004 that KSKB republishes no one's text. **Context.** On 21 October 2026 the proposal to Henk van Cann and Kor Dwarshuis ([[300-Research/310-KERISuite/docs/curator/kerisse-note-draft-2026-10|kerisse-note-draft-2026-10.md]]) promised that the text of the KERISSE dataset (KERISSE is the KERI Suite Search Engine) would stay on Mark's machine and that nothing of theirs would be republished; [[300-Research/310-KERISuite/decisions/0002-2026-10-15-database-corpus-roles-reader|ADR 0002]] and [[300-Research/310-KERISuite/decisions/0004-2026-10-21-licenses-and-no-lock-in|ADR 0004]] said the same, and ADR 0002's amendment of 2 November restated it for a cloud working copy. Henk's answers of 29 October consented to the fetch for open-source use, named KERISSE's own terms as Apache 2 for the code and CC for the content, and asked for a normal acknowledgment; the README of the KERISSE repository (read on GitHub on 3 November 2026 and filed verbatim in the fetch runbook's record) says the whole dataset is uploaded so that it "stays available as a whole (e.g. for AI training)". None of that required the promise; it was KSKB's own, and nothing in the consent asked for it. On 3 November Mark set the stance: kskb.dev is an independent site, as free as the sources' actual terms allow and not constrained by concessions the terms do not require, a front end to KERISSE's content on his own terms that goes beyond search (abstracts, levels, organization by document, source and reader), sharing with the Foundation by choice. The same day the licenses of all 77 sources in the dataset were read, from each repository's license file on GitHub, each site's terms page, or the owner's word; the table `LICENSES` in [[300-Research/310-KERISuite/scripts/corpus_kerisse.py|corpus_kerisse.py]] records each reading. The dataset's 5,107 documents fall into seven classes: | Documents | Sources | License, and where it was read | What the license allows | |---:|---:|---|---| | 3,916 | 35 | Apache 2.0: the WebOfTrust repositories, KERIDoc, WOT-terms (whose license file covers software and documentation), Samuel Smith's papers, the Foundation's locksmith, the ToIP code repositories, the Read the Docs sites | reproduction and redistribution, with the license and the attribution notices kept | | 605 | 7 | OWFa 1.0, the Open Web Foundation Final Specification Agreement: the ToIP specification repositories | reproduce, prepare derivative works of, distribute, with attribution to the specification | | 25 | 1 | unclear: the archived ACDC specification repository, whose license file only points to the current repository's contribution guidelines | treated as no license until read | | 245 | 2 | CC0 (Creative Commons Zero, the public-domain dedication) unless a document is marked otherwise: the site and PDFs of GLEIF, the Global Legal Entity Identifier Foundation, under its website terms | anything | | 170 | 1 | CC BY-SA 4.0: the eSSIF-Lab framework site | reproduction with credit; an adaptation carries share-alike | | 114 | 4 | CC, version not named: the KERI Foundation's site, the KERIConf page, the two subtitle repositories, which carry no license file; Henk's word only | depends on the version; abstract until it is named | | 32 | 27 | none found: 26 single blog posts and the handmade PDFs | the author's copyright: metadata, abstract, link, a short marked quotation | Henk's consent covers the scrape of the Foundation's server. What may be shown of a text is set by that text's own license, the same whether it came through KERISSE or straight from its source. The Slack archive stays outside every class: people's messages, with no grant from them. A second reading was by kind. Of the 5,107 documents, 3,507 are not reading material: 1,792 source-code and configuration files, 1,384 issues and pull requests, 268 build artifacts and binaries (source maps, bundles, icons, libraries) and 63 other repository files. The 1,600 prose documents are 642 glossary terms, 397 web pages, 164 specification files, 161 repository prose files, 119 PDFs, 83 KERIDoc pages and 34 subtitle files. The text worth showing is the prose; the rest is indexed with a link. **Decision.** 1. **A license per source, read and recorded.** [[300-Research/310-KERISuite/data/kerisse-sources.yaml|kerisse-sources.yaml]] carries per row `license` (an SPDX identifier, the Software Package Data Exchange's list of license ids, or OWFa-1.0, CC-unversioned, unclear or none), `license_read` (where and when it was read) and `show` (`text` or `abstract`). The master is the `LICENSES` table in the fetch script, first match wins, and a source no row matches is `unknown` and `abstract` until someone reads its terms. The database check reports an unread source as a notice. 2. **A kind per document, from its url.** Web page, PDF, glossary term, KERIDoc page, specification file, repository prose file, subtitle file, issue or pull request, repository page, code or configuration file, other repository file, build artifact or binary. The first seven are the prose kinds. 3. **Three outcomes.** *Text:* a prose document of a source whose show is text is shown in full, rendered from the corpus, under a credit line that names the source, its address, the license and KERISSE, with the notices the license asks for: the Apache license and any NOTICE file kept by reference to the repository; the specification's name and the copy's date for OWFa 1.0; share-alike carried on an eSSIF-Lab text; a CC BY credit, link and statement of changes when a CC BY source is added. *Abstract:* every other prose document, and every code file and artifact, is shown as metadata, KSKB's own abstract, a link to the original, and at most a short exact quotation marked as such. *None:* the Slack archive, and anything a person asks to have left out, is indexed on this machine only. 4. **Conversation is never republished.** Issues, pull-request threads and discussions are people's comments, not contributions licensed with the repository. They are indexed with title, state, author and link, like the Slack archive, whatever the repository's license. 5. **KSKB's own writing stays CC BY 4.0** in every class (ADR 0004, decision 2). An abstract in KSKB's words is KSKB's, not an adaptation of the text it describes. 6. **The corpus and the databases.** `.corpus/` stays gitignored and is never committed. [[300-Research/310-KERISuite/scripts/build_db.py|build_db.py]] with `--text` writes `build/kskb-text.sqlite`, one row per document whose show is text, rendered to Markdown with its credit, and a full-text index over it. kskb.dev serves it as a second database beside the metadata copy; it travels as a build artifact and is never committed to the vault or carried by an export (the ks profile excludes `build/`). `--text-all` writes the local text database of every prose document, for reading and abstract-writing on this machine, and it never leaves it. 7. **The open questions, and who answers them.** Henk van Cann names the Foundation's Creative Commons version, asked on the call of 9 November. A blog post's author may be asked for permission to show the post, and the answer is recorded on the card. A GLEIF document that carries a restriction of its own is moved to abstract when found. 8. **Said plainly to the Foundation.** The reply to Henk's answers (v0.2 in the proposal note) says that the plan has grown from a reader's index to a public site, and states this rule in one paragraph, before the call. **Consequences.** - ADR 0002, decision 2 and its amendment of 2 November, is amended: the corpus is still held locally, and what the shared copies and the site carry is set by this record, not by where the text sits. ADR 0004, decision 3, is replaced by decisions 3 to 5 above. - The manifest's `[license]` table, the folder's context file, the user guide, the core-group README, the reader's About page and the mockup's License page restate the rule. - The first build, 3 November: 1,468 documents, 16.7 million characters, 41 MB with the index. By license, 892 under Apache 2.0, 245 under CC0, 170 under CC BY-SA 4.0, 161 under OWFa 1.0; by kind, 642 glossary terms, 303 web pages, 161 specification files, 160 repository prose files, 119 PDFs, 83 KERIDoc pages. The 114 Foundation documents, the 32 unlicensed pages and the 3 archived ACDC files wait at abstract. - The grouping folder for Kor and Henk carries each item's kind and each source's license, which also flags the 268 build artifacts and 1,792 code files that KERISSE indexes as if they were pages. - The promise of 21 October is superseded, and said so in the reply before the call, so that nothing changes quietly. **Alternatives.** Keeping the blanket rule: safe, and it forgoes the text of nine in ten of the dataset's prose documents, which their licenses make free to show. Showing everything: breaks the unlicensed authors' copyright and the norm that conversation is not republished. Asking the Foundation's permission for the whole: not theirs to give for the upstream sources, and not needed for the open ones.

Links from other tables