Evidence
Every claim on this site,
and where it's proven.
Every claim on this site was checked against the code behind it. Live links open the proof; the rest is a walkthrough away.
The limits, stated plainly
- Fallback answers are flagged. IT Knows It marks its guarded known-knowledge fallback in the response and keeps it out of the cache. Grounded-with-flagged-fallback is the honest claim.
- Deploys are gated on health checks and smoke tests. The 169-prompt harness runs against production and candidate builds, with logged history.
- Humans stay in the loop. Admin review gates, clarification turns, and opt-in cloud features are built in where they matter.
- Each proof covers its own ground. The NYU numbers prove program delivery; the code proves the intelligence engineering.