Skip to content
Banford AI OmniSight

METHODOLOGY

How OmniSight earns a public claim

Collection success and editorial truth are different gates. OmniSight keeps them separate.

1. Observe

Collectors preserve point-in-time records and source timestamps.

2. Validate

Schema, freshness, duplication, and source health checks run before export.

3. Derive

Deterministic metrics and anomaly context are computed without changing the underlying evidence.

4. Challenge

Every interpretation receives a counterpoint and evidence linkage.

5. Approve

A human reviewer must approve wording before a canonical story is public.

PUBLICATION GATE

Minimum evidence

  • Metadata completeness is disclosed as a heuristic only; it cannot establish truth, relevance, or materiality.
  • Topic relevance, materiality, and event specificity must be verified against the actual evidence, not inferred from a collector name.
  • Independent evidence is counted by the actual upstream source host, with an exact item URL rather than a search, list, or API-query endpoint.
  • Current evidence, historical backfill, and raw collector observations remain separate in every claim and review record.
  • Every number must reconcile to deterministic evidence IDs, filters, timestamps, and a public-safe source bundle.
  • Original analysis, counterpoints, sensitive-topic review, and image relevance must pass before indexing.
  • Human approval is bound to an immutable content hash; changed evidence or wording requires a new review or explicit correction.

DATA USE

Derived metadata, source-specific terms

OmniSight publishes public-safe derived monitoring metadata and links back to the originating sources. Source records remain subject to each upstream publisher's terms and licenses. Unless a dataset page states otherwise, BanfordAI does not grant a new license over third-party source records. Cite the canonical OmniSight page and the linked primary sources, and do not present monitoring volume as impact or causality.