Sourcing & verification

What counts as evidence on AI Orbit.

Verification is not one universal badge. The source, claim type, freshness and measurement context all matter.

Source hierarchy

AI Orbit prefers the most direct evidence appropriate to the claim. An official provider page is usually stronger for a current price or product policy, while an independent benchmark source may be more appropriate for a performance result.

Official / primaryProvider documentation, pricing pages, release notes, repositories, policy pages and first-party announcements.

Measured / researchBenchmark datasets, research papers or structured test records with identifiable methodology.

Reliable secondaryReputable reporting or analysis used when it adds context that is not available from a primary record.

Community / opinionUseful for experience signals, but not a substitute for source-backed factual verification.

Tool profile verification

Tool pages can contain source-backed capabilities, use-case fits, lifecycle status, technical facts, privacy/security notes and pricing records. A “pending” state means the field exists or is relevant but has not met the configured verification condition. It is not automatically a negative judgment about the product.

Pricing verification

Pricing should preserve the plan, billing period, currency, unit and source context. A current price can become stale quickly, so freshness and verification dates are displayed where records support them. Custom or enterprise pricing should remain “custom” when no public numeric rate exists.

Plan contextFree, monthly, annual, usage-based and custom plans should not be collapsed into one number.

FreshnessVerification dates help readers judge how recently a commercial fact was checked.

Change historyHistorical snapshots can be separated from the currently stored plan state.

Benchmark evidence

Verified benchmark records should preserve the benchmark identity, score, source, test date and semantic class. Scores from materially different test types should not be blended as if they measured the same thing. Composite scores are comparative interface metrics, not universal product-quality claims.

Read benchmark methodology

News relevance & verification

AI news ingestion can use source identity, title/body relevance, entity context, duplicate detection and editorial override controls. A company or product name by itself should not make an unrelated story “AI news.” AI-generated summaries are treated as summaries and the original source remains the factual reference point.

Confidence scores

Profile confidence summarizes evidence coverage, verification and freshness for the stored record. It should not be read as a product-quality score. A low confidence score can mean incomplete evidence even when the product itself is excellent.

InterpretationConfidence = evidence coverage + verification + freshness signalsThe exact implementation can vary by record type and available fields.

Freshness & limitations

AI products, prices and policies can change after verification. Dates and source links are therefore part of the evidence. If you find a newer official source or a stale record, report it so the structured data can be reviewed.

Report stale data Corrections Policy