01 · Learn
Separate knowledge types
Customer assertions, inferred capabilities, preferences, evidence, confidence, and confirmation state remain distinct and auditable.
Search methodology
Company context in. Explainable SAM.gov matches out.
The agent builds a versioned internal search plan from customer evidence, confirmed facts, labeled inferences, corrections, and feedback. Customers do not have to build filters.
02 · Search
Use structured and semantic signals
Each run records its plan version, filters, semantic query, ingestion watermark, candidate counts, scoring components, exclusions, duration, and errors.
03 · Explain
Never hide the source
Every recommendation includes a concise match reason, an explainable relevance and confidence label, and a link to the authoritative SAM.gov notice.
Quality boundary
No weak-result padding.
If no new record clears the relevance threshold, the brief says so. Feedback changes preferences gradually; a single downvote never silently becomes a permanent exclusion.