Trajectory

A customer requests their records to understand how their data is being used. The request is authorized and queued, but access is revoked before the records are delivered. Should the system honor authorization at request time, or recheck it at delivery time? Trajectory is a decision-based workbench that makes this choice explicit and connects it to a working change. In the demo, the owner selects delivery-time authorization. Meta Muse synthesizes a bounded policy; an independent checker evaluates the original behavior, an intentionally wrong fix, and the generated candidate against the same examples. The checked source is applied in a private workspace, and a new job replays the scenario. After revocation, the replay blocks delivery and records zero exports. Receipts connect the decision, source, checks, application, and observed result. A separate request-time run preserves the originally authorized request and records one audit and one export. An interrupted authorized delivery is recovered with one audit and one export. The demonstration shows how an owner can move from an ambiguous design decision to an inspectable, independently checked outcome without writing code. For the hackathon we built the bounded records-delivery companion workflow, policy synthesis, independent checks, checked application, replay, durable audit and recovery receipts, and the native workbench flow. The project builds on prior BMO research in decision-guided synthesis and independent verification and an existing native workbench. The repository identifies the event-specific companion separately from the inherited workbench.