Capability Gaps Surfaced by the First Triage Run

Run manifest with full context: clients/reach-edu/corpus/inbox/_triage-runs/2026-07-24_1.md. Sibling issue: [[Parent-Child-Nested-Organizations-Not-Modeled]] (the one gap big enough for its own file).

1. No org-side observation capability

person.add_observation exists; organizations have nothing. The operator’s “flag it association and network” / gov-entity flags had to be direct SurrealDB writes mirroring createObservation(). Wants an organization.add_observation twin.

Why it’s worse than inconvenient — the typing hazard: the direct writes initially stored subjects as string-typed record ids (organizations:⟨uuid⟩) while org rows use uuid-typed ids (organizations:u"uuid"). The observations were self-consistent but did not join to their org rows — silently. Discovered and repaired 2026-07-25 (9 observations re-typed). This is exactly the class of bug the capability wire exists to prevent; every future direct write risks reintroducing it.

2. conventional_name has no edit affordance in the entity card

resolver.update_org sets it (and does slug renames, preserving old slugs as aliases), but the UI’s chip editors cover only aliases/domains (per Entity-Card-Edit-And-Remove-Affordances). Related: the create flow mints all name fields from one string (“Stanford-PACS” became both complete_name and conventional_name), so every UI-created org needs a post-create enrichment pass. The ratified naming model (slug derived / conventional_name / complete_name / greedy aliases) lives in the triage skill.

3. source.add has no slug override for fetch-hostile URLs

The capability Jina-fetches metadata to derive the source slug/filename. When the fetch hits a wall, the slug is garbage: a ccdaily.com article slugged https-match-adsrvr-org-track-cmf-google (ad-tracker redirect), a bizneworleans.com article slugged 403-forbidden. Disk files were renamed by hand; DB source_usages.corpus_path still records the junk paths for those two (workforce-development sources). Wants a slug/title override arg, or a post-hoc path-patch capability.

4. source.fetch clobbers merged canonical bodies

The triage mechanics ruling (inbox-file-is-canonical) merges the already- fetched inbox body over the metadata-only stub, but the DB registry still says metadata-only. Running source.fetch on such a source would refetch and overwrite the canonical verbatim body. Interim rule: never run it on merged sources. Real fix: a registry-status patch, or a source.add variant that accepts a provided body (which would also fix half of gap 3).

5. No org delete/merge capability

The duplicate bhef org row could not be removed over the wire. Resolution used (2026-07-25, non-destructive, reversible): corpus entries moved to the keeper via remove/add; keeper aliased BHEF/bhef; orphan row updated in place — slug renamed to bhef-merged-duplicate, merged_into pointer set to the keeper, client_access cleared (original value recorded in merged_note) so it vanishes from workspace reads without losing data. Wants a first-class organization.merge that does exactly this.