Every knowledge unit on acta.today is produced through multi-model deliberation: four frontier AI models independently analyse a question, cross-critique each other with specialised roles (verifier, devil's advocate, synthesiser, clarity editor), and a synthesis engine extracts where they agree and disagree.
Every model response is cryptographically signed with an Ed25519 receipt at the moment of generation. The full deliberation chain is independently verifiable.
Automatic re-deliberation. When the model roster changes (a new frontier model enters the top 8), knowledge units produced with displaced models are flagged as stale and queued for re-deliberation. Previous versions are preserved with their original receipts.
Freshness expiry. Each knowledge unit has a freshness window of 90 days. After this, the unit is eligible for re-deliberation even if the roster has not changed.
Community reports. Readers can report factual errors or outdated information via hello@acta.today. Reported units are prioritised for re-deliberation.
All versions of a knowledge unit are preserved. When a unit is re-deliberated, the new version links to the previous one. Original receipts and deliberation transcripts are never deleted or modified.
We do not manually edit model outputs. We do not suppress disagreement. If frontier models genuinely disagree on a topic, that disagreement is surfaced as a feature. Honest uncertainty is more valuable than false consensus.
Every knowledge unit can be independently verified at acta.today/v/{ku-id}. For command-line verification: npx @veritasacta/verify check {ku-id}