Package Overview¶
bijux-canon-index exists to make retrieval behavior explicit, replayable, and reviewable. It turns prepared ingest output into embeddings, index state, and retrieval results that downstream packages can inspect instead of merely trust.
Role Model¶
flowchart LR
prepared["prepared ingest output"]
index["retrieval and indexing behavior"]
results["replayable retrieval results"]
downstream["reason, agent, and runtime consumers"]
prepared --> index --> results --> downstream
This page should let a reader picture index as the package that owns retrieval semantics in the open. The result is not just search output; it is search behavior that later packages can replay, compare, and review.
Boundary Verdict¶
If the work changes how search is executed, replayed, compared, or exposed as retrieval output, it belongs here. If it changes source preparation, claim meaning, or governed run policy, it does not.
What This Package Makes Possible¶
- retrieval behavior becomes a named contract instead of an accidental consequence of backend code
- provenance and replay stay attached to search results that later packages rely on
- index-specific search assumptions stop leaking into reasoning and runtime layers
Tempting Mistakes¶
- treating vector execution as infrastructure plumbing instead of package-owned behavior
- burying caller-facing retrieval differences inside plugins or backend adapters
- using runtime authority to paper over unclear retrieval semantics
First Proof Check¶
packages/bijux-canon-index/src/bijux_canon_indexfor retrieval ownership in codepackages/bijux-canon-index/apisfor tracked caller-facing schemaspackages/bijux-canon-index/testsfor replay and provenance evidence
Design Pressure¶
The pressure on index is to keep retrieval logic visible enough that later packages never need to guess what happened inside search. If retrieval policy spills into adapters or downstream code, the package boundary stops paying off.