modelProvider/modelVersion: local / gemma4:e4b
Lane: local_report_generation
OpenAI content/report lanes require explicit gates and use store:false; no-call previews write local artifacts only.
Generated reports inherit cited report text, source facts, packet evidence, saved source_ref claim policy, and /v1/sources source-fact health: sourceFactHealth.total > 0, sourceFactHealth.claimBearingPolicyRows > 0, parser-lag/future-row/embedding-backlog/degraded-run checks, and no blocking claimWarnings.
Generated report wording is reusable only inside cited report text, cited packet evidence, or saved source_ref claim policy.
Broad source-level claims require sourceClaimAudit review, /v1/sources claimBearing=true, sourceFactHealth.total > 0, sourceFactHealth.claimBearingPolicyRows > 0, sourceFactHealth.parserLagHours within maxParserLagHours, sourceFactHealth.futureEventsBeyond7d=0 for current road claims, no pending/failed embedding backlog for semantic-only claims, no degraded-run caveats, and no blocking claimWarnings.
Raw-current, stale, no-row, scheduled, planned, retired, unknown, planned/future, parser-lagged, parser-incomplete, degraded-run, and embedding-backlog rows remain context only unless cited evidence says otherwise.
Do not infer causality from co-occurrence; weather, road, policy, and wait overlap stay contextual unless evidence proves cause.
Persisted with report: no