Methodology

Source acquisition

Each connector fetches official public data within documented authentication, pagination, date-range, and rate-limit requirements. Fetches use bounded retries. Malformed records are quarantined instead of failing an entire run.

Provenance and snapshots

Every normalized record retains its source name, source record identifier, source URL, source timestamps, observation times, ingestion time, and checksum. Permitted raw payloads are retained in an audit table or object storage. A new checksum creates one version snapshot; repeated payloads do not create duplicate history.

Normalization

Agency names, locations, notice types, set-asides, dates, NAICS codes, and product service codes are mapped to consistent fields. Important eligibility, contact, and deadline facts remain faithful to the source. Missing fields display as “Not stated.”

Industry classification

Rules consider source NAICS and product service codes plus title and description language. The eight user-facing hubs improve discovery but never replace official classifications. Low-confidence records go to Other for later review instead of being forced into a category.

Deduplication and indexability

Source name and source record identifier form the primary deduplication key. Pages missing a usable title or source, obvious duplicates, and records with too little public information are not indexable. Expired pages stay indexable only when their archival or award context remains useful.

Alerts

Search matching runs against normalized filters. Alert event keys combine the saved search, opportunity, event type, and record checksum. Delivery keys combine the event and recipient so a retry cannot send the same email twice.

Historical awards and fit explanations

USAspending context is shown only when identifiers or high-confidence fields support the match. The evidence and confidence explanation are stored. Fit scores, when introduced, list the contributing factors and never conceal official eligibility rules.