An interactive lab for deterministic AI agent authorization A public interactive lab now lets users run documented, local JavaScript policy-and-evidence functions that authorize or block a proposed AI-agent operation before any synthetic execution effect occurs, with each evaluated layer returning a reason and any mandatory failure stopping the remaining checks. The demonstration includes an optional fifth step implementing a simple two-export demonstration budget, which the authors state is not the patented or unpublished Session Risk Memory algorithm and does not establish incremental security benefit. The lab writes only to an in-browser synthetic effects ledger, includes no real credential attestation, encrypted claims, proprietary authorization engines, real tool sandbox or production enforcement, and its authors are seeking independent reviewers, applied research collaborators, AI security teams, and organizations interested in controlled pilots or technology licensing. Deterministic evaluation Every run executes a documented, local JavaScript policy-and-evidence function for each stage. Change the fields and the decision can change. See how independent, deterministic checks can authorize — or stop — a proposed AI-agent operation before a synthetic execution effect occurs. Choose a synthetic request, change its evidence, and run the same public rules each time. Each evaluated layer returns a reason. Mandatory failure stops the remaining checks. No synthetic operations recorded yet. Run a scenario to inspect the result. The optional fifth step implements a simple two-export demonstration budget . It is not the patented or unpublished Session Risk Memory algorithm, and it does not establish incremental security benefit. ALLOW appends an entry to the in-browser synthetic effects ledger. BLOCK appends no execution effect. Nothing is written to bank systems or remote services. No real credential attestation, encrypted claims, proprietary authorization engines, real tool sandbox, or production enforcement is included in this public demonstration. Seeking independent reviewers, applied research collaborators, AI security teams, and organizations interested in controlled pilots or technology licensing.