1. Source registration
Every adapter has a source owner, access method, policy review, request budget, field map, retention rule, and permitted product uses. Official APIs are called rather than scraped where available.
2. Immutable evidence
The original response and its retrieval metadata are retained before parsing. Derived values point back to evidence or an integrity hash.
3. Three timestamps
- Event time describes when the underlying event occurred or took effect.
- Publish time describes when the source made it available.
- Fetch time describes when SentientData retrieved it.
4. Entity resolution
Exact official identifiers, domains, and product identifiers resolve deterministically. If identifiers point at multiple organizations, ingestion stops for review. Probabilistic matches are quarantined and never silently merged.
5. Review and publication
Deterministic, high-confidence official registry changes may auto-publish under a versioned rule. Ambiguous organization matches, website inferences, and material score changes require review or a measured canary.
6. Score boundaries
Identity/compliance, operational activity, and digital readiness are separate. Technology choices can describe readiness or sales fit but cannot establish legitimacy. Payment never changes a score. Missing evidence remains unknown.
7. Corrections
Corrections preserve the original observation, submitted evidence, decision, and material downstream update. See the corrections process.