Agent Lab — Working in unison

Work together through everyday goals, questions, boundaries and corrections. Inspect one interaction record through conversation, review and research views, with a separate preserved mechanics demonstration.

Part of the Trust Scalar research hub

Begin with a guided browser-only example or explicitly consent to a text-only live model session. Goals, preferences, boundaries, questions and corrections guide collaboration. No outside tools or irreversible actions are connected.

Work together, understand what changed and inspect research through the same event record. Proposals, approvals, draft applications and human assessments remain distinct. Saved guidance is inspectable and removable; recording a correction does not demonstrate transfer.

Observing collaboration without an added numerical controller is different from engineering a review-based permission gate. Both retain common safeguards. Predictive fit would not establish a unique underlying architecture. A local SHA-256 lock is not a public preregistration.

Preserved mechanics demonstration

The separate rule-based simulator retains A/B/C comparisons, C−B, B−A and C−A, held-out cases, replay, questionnaire-removal tests and exports. Its programmed results are not evidence from live agents.