Trust policy

Editorial Policy

Novamente AI Lab publishes practical, evidence-first content for reliable AI workflows, but it does not currently run first-party model evaluations.

Benchmark method

Benchmark pages synthesize and cite public third-party leaderboards. Every public figure should be linked to its source and dated in the copy. The site does not publish a Novamente ranking until a dated first-party run log exists for that page.

Review rules

• Current claims about model performance, pricing, privacy, or product behavior must be checked against a live source before publication.
• Tool pages must state the task fit, data boundary, and what verification work remains with the operator.
• Historical archive pages must distinguish source-backed history from modern interpretation.
• Articles should end with a checklist, template, workflow, source packet, or other reusable artifact.

Corrections and updates

If a benchmark board, provider policy, or model behavior changes, the page should be updated with a new date or narrowed with a hedge. Stale confidence is worse than a smaller claim.