A derivation explains itself down to the sentence

[WaylandYang]

B1 of 0016 (R2 in 0002). A derived fact could show its direct premises as text; now it explains itself down to the sentence. reasoning::proof returns the derivation plus its premises in derivation order, each carrying its evidence rows; GET /kbs/{id}/derived/{id}/proof serves it; in the entity panel a derived row expands into the chain, each step a triple with its quotes linking into the document.

It is a chain, not a tree: fact_derivations records only asserted premises (derivation input excludes derived facts), so "recursive expansion" flattens to the ordered premises and their evidence. A premise that was since retracted is listed and marked, not hidden — the derivation falls with it, but "what it rested on at the time" stays readable, which is what the record axis is for. A derivation that no longer exists yields proof: null, and the panel falls back to the text premises it already had.

Decision records: 0002 R2 marked done with a revision on the chain-vs-tree shape; 0016 B1 marked done; index updated.

Checks: clippy clean, pnpm typecheck clean, new store test a_proof_reaches_the_sentence (order, leaf quotes, retracted marking) green alongside derived_facts_are_second_class; verified in the browser on a scratch base (transitive part_of chain and a sub-property derivation whose step links to the source sentence).

🤖 Generated with Claude Code