docs(wikidata): PRD-0011 + ADR-0016/0017 — Wikidata as a first-class SP42 target#105
Merged
Conversation
Establishes Wikidata as a patrol target and a referenced source/sink of facts for the citation work. Splits the design along the reuse-by-design rule: a shared platform entity read mechanism (revision read + EntityDiff + content-model routing/gating, spawned as ADR-0015) versus thin Wikidata-native workflows. MVP is the patrol read experience (entity diff render, scoring gated off, testwikidatawiki) with the use-case family — book/entity enrichment, citation→Wikidata facts, Wikidata→sources — sequenced as follow-ons. Continues ADR-0014. Adds the wikidata domain README and indexes it in docs/README and docs/domains/README. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012u5MVp3vAN44KhsVg3vroY
Fold in the six design-review decisions and convert Open questions to Resolved: Q1 hybrid/defer (read mechanism -> platform, patrol-Wikidata -> patrolling, no new crate; write-lane home deferred); Q2 read production wikidatawiki, gate writes to testwikidatawiki; Q3 chronological over a bot-filtered stream (rcshow=!bot); Q4 full statement depth (property/value/qualifiers/rank/ references) with the never-a-no-op invariant; Q5 citation->facts first; Q6 ADR-0015 read now + a separate ADR-0016 write contract. Updates the header, Proposal, DoD, and the domain README to match. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012u5MVp3vAN44KhsVg3vroY
…tement-write contract) Draft both ADRs PRD-0011 spawns, up front, so the PRD's forward references are concrete before review. ADR-0015 (read mechanism, platform): content_model as an additive per-revision fact (default wikitext) with routing keyed on the revision — never the wiki id, since wikidata.org has wikitext pages and Wikipedias have JSON; entity revision read via the Action API prop=revisions; a new EntityDiff sibling to StructuredDiff diffing statements at full depth (property/value/qualifiers/rank/ references) with a never-a-no-op invariant; a ContentDiff sum for routing; feature gating by content model as a separate axis from the OAuth-derived capability profile; bot filtering reuses the existing rcshow=!bot flag; scoring skipped, not faked. ADR-0016 (write contract, platform): reuses ADR-0010 propose/confirm, swapping the wikitext node locator for entity id + lastrevid drift and the payload for a grounded, referenced statement; structured triples gated by ADR-0007; sibling apply route; per-operator Wikidata auth; testwikidatawiki write gate. Reconciles the PRD-0011 and wikidata README ADR references and indexes ADR-0013..0016 in the platform README. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012u5MVp3vAN44KhsVg3vroY
…#99) PR #99 already claims ADR-0015 (references/adr, rules-compliant fetch edge). ADR numbering is global, so the Wikidata entity-read and statement-write ADRs move to the next free numbers: read -> ADR-0016, write -> ADR-0017. Renames the two files and updates every cross-reference in PRD-0011, the wikidata README, and the platform README index. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012u5MVp3vAN44KhsVg3vroY
…P surface) Cross-check against PR #103's verify_wikidata_statement, which is the first Wikidata read in the codebase (Special:EntityData + statement/P854 parser) but lives in the sp42-mcp shell. Reconcile: - ADR-0016 Decision 2: adopt and PROMOTE #103's entity/statement parser out of sp42-mcp into platform so the MCP verb, patrol entity diff, and the write lane share one Wikidata read model; endpoint (Action API prop=revisions vs Special:EntityData?revision=) is an impl detail on shared, endpoint-agnostic parsing. Add a Relation-to-#103 note and coordination point. - ADR-0017: record that #103's verb is the exact inverse of the statement-write contract (verify existing reference vs propose new referenced statement); same statement + P854 model and ADR-0007 grounding, two ends of one loop. - PRD-0011: Wikidata->sources reuses #103's statement/reference read (note the external Wikimedia-DE Wikidata MCP as an alternative read source); citation-> facts noted as ADR-0017-governed and #103's inverse. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012u5MVp3vAN44KhsVg3vroY
…#103 Add a pre-implementation design-plan sketch of the typed Wikidata read model (Entity/Statement/Snak/WikibaseValue/Reference, endpoint-agnostic parse, label resolution, statement rendering, EntityDiff/ContentDiff, StatementRef and the ADR-0017 StatementProposal) that lives in platform (sp42-platform::wikibase) rather than the sp42-mcp shell. Includes a line-by-line mapping from PR #103's current wikidata.rs onto the shared API (get_json/parse_statement/render_value/ extract_ref_url/label_of -> the typed model), the additive delta patrol and the write lane need, and four open points to settle with #103 (ContentDiff home, diff fetch endpoint, merge order, renderer fidelity). Link it from ADR-0016. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012u5MVp3vAN44KhsVg3vroY
…nd PRD-0011 Research check on existing Open Library <-> Wikidata data sharing surfaced two things worth encoding before review: - Existing sync (do not duplicate): P648 links OL<->Wikidata; OL runs an official Wikidata Integration (author bios/photos/awards in, OL IDs out) plus cdrini/openlibrary-wikidata-bot for identifier sync. SP42's distinct value is the grounded/referenced-statement layer, answering the standing WikiProject complaint that most bots add unsourced statements. Recorded in ADR-0017 Relation + a PRD-0011 risk (with a follow-up to mirror it into PRD-0009). - FRBR work/edition split (real modeling constraint): ISBN/publisher/pagination belong on the edition item and are sourced there, not the work; a past bulk import mis-placed ~40k edition IDs on non-edition items. ADR-0017 Decision 8 now targets the edition a citation identifies and declines when no edition item exists rather than creating one; PRD-0011 risk records it. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012u5MVp3vAN44KhsVg3vroY
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 99f14c6960
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
…review) Codex review correctly flagged that an action binds to the (wiki, rev_id) of the change — the workbench copies the event's wiki_id into the action request and the server acts on exactly that wiki — so "read production wikidatawiki, write to testwikidatawiki" is impossible as worded: a production revision does not exist on the test wiki, and patrol/rollback cannot be redirected cross-wiki. Reword Q2 (MVP proposal, DoD, Resolved) to the coherent form that preserves the intent: production wikidatawiki is review-only (real change volume validates the entity diff at zero write risk); the patrol/rollback action loop runs on testwikidatawiki in place (a test change reviewed and acted on there) as the live write gate; production actions stay disabled until proven. This matches the shipped Wikipedia posture (production reviewed, edit/action gate on testwiki). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012u5MVp3vAN44KhsVg3vroY
tieguy
added a commit
that referenced
this pull request
Jul 2, 2026
PRD-0011 is claimed by the Wikidata target PRD (#105) and ADR-0015 by the fetch-edge ADR already on main; the citation-insertion draft this PRD depends on becomes PRD-0012. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Design docs (no code) establishing Wikidata as a first-class SP42 target — not
just a book-metadata source, but a patrol target and a bidirectional bridge with the
citation work. SP42 already resolves
wikidatawiki/testwikidatawiki(ADR-0014)but can't review them: a Wikidata revision is entity JSON, so the diff view renders
raw JSON and every wikitext signal misfires.
patrol-read MVP, with the use-case family (book/entity enrichment, citation→Wikidata
facts, Wikidata→sources) sequenced as follow-ons. Six design decisions resolved
in-doc.
content_model(routing keys on the revision, never the wiki id), entity revision read, a new
EntityDiff(full statement depth, never-a-no-op),ContentDiffrouting, andcontent-model capability gating.
ADR-0010 propose/confirm with an entity+
lastrevidlocator and a grounded,referenced statement payload; respects the FRBR work/edition split.
2026-07-01-wikidata-read-model.md— the shared platform Wikidataread model and a line-by-line convergence mapping from PR Citation verification over MCP — read-only agent surface (MVP) #103's
wikidata.rs.New
docs/domains/wikidata/domain, indexed indocs/README.mdanddocs/domains/README.md.Type
Validation Required
Docs-only change (Markdown under
docs/); no source, build, or test surface touched.# nothing to run — no code changedRisk
Two ADRs are included here (ADR-0016, ADR-0017), both
Proposed. No further ADR isrequired to merge these drafts.
Notes For Reviewers
Coordination dependency — PR #103 (PRD-0010, MCP surface). #103 ships the first
Wikidata read in the codebase (
verify_wikidata_statement,sp42-mcp/src/wikidata.rs)but in the
sp42-mcpshell. Per the reuse-by-design rule, ADR-0016 promotes thatentity/statement parser into platform so patrol, the MCP verb, and the write lane
share one read model. The design plan gives the line-by-line mapping; #103's own design
plan already anticipated a platform home for the renderer, so this is convergence, not
conflict. Whichever of #103 / this lands first, the other consumes or refactors onto
the shared module — #103's public verb contract (PRD-0010) is unchanged either way.
ADR-0017 is the exact inverse of #103's verb (propose a referenced statement vs. verify
an existing one).
Numbering. #99 holds ADR-0015 (fetch edge), so these took the next free global
numbers, 0016/0017. PRD is 0011 (0010 is #103's).
Ecosystem due-diligence (baked into the docs). OL and Wikidata already sync (P648;
OL's official Wikidata Integration;
cdrini/openlibrary-wikidata-bot). SP42 does notrebuild identifier sync — its value is the grounded/referenced-statement layer, which
answers the standing WikiProject complaint that most bots add unsourced statements. And
ADR-0017 respects the FRBR work/edition split (ISBN/publisher on the edition item,
sourced there; decline rather than create a missing edition — avoiding the ~40k
mis-placed-edition-ID class of bulk-import error). A follow-up note tracks mirroring the
de-dup point into PRD-0009 (PR #104).
Most worth scrutiny: ADR-0016 Decision 5 (content-model capability gating as a new
axis on the profile) and the PRD's resolved Q1 (hybrid/defer on domain placement — no
new crate for the MVP). Everything is
Draft/Proposed; this PR exists to get thedesign reviewed before code.
🤖 Generated with Claude Code
Generated by Claude Code