Skip to content

Doc progressive disclosure: L0/L1/L2 excerpt levels + retrieval trajectory explain #85

@aurexav

Description

@aurexav

Context\n- The product goal is token-efficient memory usage: return compact results first, then expand/hydrate evidence on demand.\n\nGoal\n- Add progressive disclosure for docs: return L0 (metadata + short snippet) then L1/L2 (larger excerpt windows) with an explicit retrieval trajectory.\n\nDeliverables\n- API additions:\n - docs_search returns L0 results with stable locators\n - docs_excerpts_get supports level=0|1|2 (bounded caps per level)\n- Trace/trajectory model for doc retrieval (stage-level explain):\n - candidate retrieval\n - filter application\n - fusion/rerank (if enabled)\n - excerpt hydration\n\nAcceptance criteria\n- Deterministic size caps per level.\n- Trajectory explain output is machine-readable and linkable to source_ref.\n\nRelated\n- #80 (observability)

Metadata

Metadata

Assignees

No one assigned

    Labels

    area:apiHTTP API service and request/response contracts.area:observabilityLogging, tracing, and diagnostics.area:serviceRetrieval logic, ranking, and request orchestration.kind:featNew capability or product behavior that is not primarily a refactor or cleanup.theme:costLatency, compute, storage, and cost controls.theme:provenanceEvidence, citations, lineage, and explainability.

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions