[gedaiu]

Brings re-dev up to upstream dev at d62e880b: 11 commits, deeplethe#1318 (name facts are found through a predicate index), deeplethe#1320 (containment matches % and _ literally), deeplethe#1264 (a token reads the graph overview and neighbourhood), deeplethe#1319 (ADR 0071), deeplethe#1299 (Markdown name variants resolve on the same paths), deeplethe#1323, deeplethe#1304, deeplethe#1305, deeplethe#1306, deeplethe#1324 (intake jobs are claimed before model work) and deeplethe#1325 (a base can follow a GitHub repository's markdown files). The dev mirror was fast-forwarded to upstream first. deeplethe#1324 and deeplethe#1325 came in as a second merge on this branch, on top of the first one at fa6cd6cc.

Conflicts. Upstream's text where only upstream changed it, ours where a fork-only change sits on top, and both where each side added a line.

• Finding a base's name facts without vectors reads only its name facts deeplethe/utopia#1318 is the same change as our Finding a base's name facts without vectors reads only its name facts (#1318, ahead of upstream) #10. name_vectors.rs and name_vectors_plan_tests.rs merged clean and are now identical to upstream's. Both migrations stay: upstream's 0115 and our 9001 are the same CREATE INDEX IF NOT EXISTS facts_live_predicate_idx, so whichever runs second does nothing.
• resolution.rs: ours for containment recall, which is Containment matches a name's % and _ literally instead of as wildcards deeplethe/utopia#1320 (our Containment matches a name's % and _ literally instead of as wildcards (#1320, ahead of upstream) #11) with Containment recall reads indexes instead of scanning the base #12's trigram rewrite on top (CONTAINMENT_SQL, contained_names, SET LOCAL jit = off), plus our Recall::Full guard from the pushed-name path. Markdown markers around a name no longer keep it from meeting the same name written plainly deeplethe/utopia#1299's Markdown folding came in around it: name is now the folded &str, so our calls pass name rather than &name. Containment matches a name's % and _ literally instead of as wildcards deeplethe/utopia#1320's like_escaped and its two tests are in place and pass.
• a_similar_name_is_proposed_not_merged.rs: ours, which keeps Containment recall reads indexes instead of scanning the base #12's mr_holmes_is_paired_with_sherlock_holmes_through_its_alias_holmes. Markdown markers around a name no longer keep it from meeting the same name written plainly deeplethe/utopia#1299's change to the_same_literal_name_is_not_proposed_twice merged clean.
• Token reads (A token reads a base's graph overview and neighbourhood deeplethe/utopia#1264): upstream's text, with the fork-only reads added to its lists: ontology (A token reads a base's ontology and a class's instances deeplethe/utopia#1265), extractions and export (A token reads a document's extractions and exports its base deeplethe/utopia#1266), read-back by external_id (A client reads a document back by the external_id it pushed it under deeplethe/utopia#1280) and chunk search. This covers the auth.rs comment, ADR 0014 decision 7 and access-and-audit.md. ADR 0014's revisions keep our 2026-10-16 line, which already counts every graph read. Upstream's line, which adds the overview and neighbourhood on 2026-10-19, is left out. graph_routes.rs was already identical, and token_read_tests.rs takes upstream's six paths.
• ADR 0054: both revision lines, ours of 2026-10-17 then upstream's of 2026-10-19 (0071).
• Second merge (Intake jobs are claimed before model work, so a sync no longer waits behind the extraction backlog deeplethe/utopia#1324, A base can follow a GitHub repository's markdown files, fetching only the files that changed since the last sync deeplethe/utopia#1325): the only conflict was CURRENT_SCHEMA_VERSION, set to the recount (107). Neither PR touches a file the fork changed apart from models.rs and utopia-server/src/main.rs, which merged clean.

Behaviour to know about.

• Markdown markers around a name no longer keep it from meeting the same name written plainly deeplethe/utopia#1299 folds **Name**, __Name__ and backticked paths inside resolve_by_name and names::record, so a pushed name (resolve_pushed_name, Recall::Spelling) is folded too. A client that pushes **Pump 1** now gets Pump 1.
• ADR 0071 (proposed upstream) makes exact-name resolution an opt-in for a statements source ({"names": "identity"}). Here every pushed name already resolves that way (RC-002, c80af41e, b7584090). If 0071 lands as written, the two will need reconciling.
• Intake jobs are claimed before model work, so a sync no longer waits behind the extraction backlog deeplethe/utopia#1324 orders claims by job_priority(kind) (migration 0116): sync, document intake and embedding kinds come before anything that calls the chat model. Our fork-only embed_names job is not on that list, so it is claimed at priority 1, behind the extraction backlog, even though it only calls the embedding model. Pushed names do not need it (they resolve by name alone), but channel-2 recall for non-pushed documents waits for the backlog to drain. Adding it to the list would take a fork migration (9003) that runs CREATE OR REPLACE FUNCTION job_priority and rebuilds jobs_priority_claim_idx. That is a separate decision for Bogdan.
• A base can follow a GitHub repository's markdown files, fetching only the files that changed since the last sync deeplethe/utopia#1325 adds a github_repo source kind that pulls a repository's markdown files, one document per path, refetching only files whose blob SHA changed. It does not interact with the fork's push sources: the read-back by external_id (A client reads a document back by the external_id it pushed it under deeplethe/utopia#1280) stays limited to api and statements, so a github_repo document is not readable that way. It is an alternative to mirroring docs through a write token's api source (A write token creates an api source as well as a statements source deeplethe/utopia#1274).
• embed_names: Finding a base's name facts without vectors reads only its name facts deeplethe/utopia#1318 is now upstream, but our embed_names job (0c56fd13), which upstream declined, stays on re-dev. Whether to drop it is a separate decision for Bogdan.

Schema version 107. Upstream added 0115 and 0116, and we carry 9000, 9001 and 9002, so there are 107 migration files. Upstream dev is at 104.

Checked on the merge against a fresh database: the migrations ran twice and the second run succeeded. cargo fmt --all -- --check and cargo clippy --workspace --all-targets -- -D warnings are clean. UTOPIA_TEST_REQUIRE_DB=1 cargo test --workspace is green: 1671 passed, 0 failed, 6 ignored (server 706 passed and 2 ignored, the store integration suite 329, store unit tests 91 with 1 ignored). deeplethe#1325 changed web files, so the web checks ran too: pnpm install --frozen-lockfile, the style guard and tsc --noEmit are clean, and vitest run passes 307 tests in 44 files.

🤖 Generated with Claude Code