Robbie now verifies how it should answer before the single AI request is sent, clustering evidence, flagging contradictions, estimating unsupported-answer risk, and optionally showing a concise trust summary in investigation workflows.
Changes in this release
- The internal investigation prompt now explicitly requires evidence-backed answers, cautious wording, and clear missing-context statements instead of unsupported claims
- Evidence is now clustered across terminal and diagnostics, Git and deployments, Docker and logs and metrics, WordPress and PHP logs, database and query failures, and connection health and service availability
- Verification warnings, contradiction indicators, and unsupported-answer risk are now computed before Robbie answers
- A concise trust summary can now appear in the Robbie reply when confidence is not high or the request is an investigation workflow
- The internal debug panel now shows verification warnings, evidence clusters, contradiction indicators, unsupported-answer risk, and trust-summary previews
[Read the full blog post →](/blog/robbie-v2-verification-trust-layer)
*Released 3 July 2026 · Category: AI*