Commit Graph
2 Commits
Author SHA1 Message Date
Kai 14c97d88cf test(curator): re-record phase 2 goldens against the capable pipeline
10/10 pass: pure-model write gate holds the no-write invariants (question/injection/destructive), session-native cross-turn back-references resolve, and agentic source extraction recovers the author:title thin-body case.
2026-08-30 07:13:24 -07:00
Kai 5de80905ba eval(golden): pin bare cross-turn back-reference ("收")
Recording for the new cross_turn_bare_backreference case: ask "is it worth
collecting?", then a lone "收" must resolve to that film (collect / movie /
巴尼的人生) and propose collect with the tmdb id -- not ask which work.
2026-08-29 15:44:59 -07:00