haiku.rag/tests/capabilities
Yiorgis Gozadinos d9f489dcc8
Deduplicate search results within one model response
Sibling searches emitted in one response overlap heavily (40.6% of
returned chunk slots on Glimmer ORB fan-out cases). A result whose
rendered evidence a sibling already showed keeps its rank slot but
collapses to a reference line, and a picture attaches once per response
keyed on (source, document_id, self_ref). Equivalence is the
format_for_agent rendering at neutral rank/total plus picture keys,
bucketed under the qualified chunk id, so another database's copy or a
different expansion of the same anchor formats in full.

Search state now commits only after formatting and image construction
succeed: a raising image build no longer leaves results citable that
the model never saw, notes evidence for them, or suppresses a later
sibling.
2026-09-02 12:29:12 +03:00
..
conftest.py replace haiku.skills with native Pydantic AI capabilities 2026-07-24 15:26:17 +03:00
test_agent_spec.py Resolve a capability's databases once, into a scope 2026-08-26 12:20:21 +03:00
test_borrowed_client.py Lend the caller's client to the capability 2026-08-18 14:58:18 +03:00
test_capabilities.py Price qa.max_searches in search units 2026-09-02 12:29:12 +03:00
test_citation_policy.py Price qa.max_searches in search units 2026-09-02 12:29:12 +03:00
test_evidence_capsule.py Finish the comment pass, and escape document fields everywhere Rich renders 2026-08-28 15:34:47 +03:00
test_evidence_ledger.py Enforce a declaration wherever there is something to declare 2026-08-13 13:39:42 +03:00
test_evidence_wire.py Price qa.max_searches in search units 2026-09-02 12:29:12 +03:00
test_search_units.py Deduplicate search results within one model response 2026-09-02 12:29:12 +03:00