The execution model
From a declared obligation to inspectable evidence.
1. Select an approved scope
A profile identifies the registered adapter, operation, expected result, required capabilities and bounds. Missing required capabilities stay UNSUPPORTED. The public demo is recorded playback, entirely separate from live execution.
2. Run through private supervision
The web request queues a durable job. A private executor reauthorizes the project/target and invokes the existing CLI through fixed arguments. On QNX, the independent supervisor delegates every potentially blocking service operation to an owned worker.
3. Inspect each obligation
Native pending, trigger, result, state, fresh-client health and cleanup observations remain distinct. The existing deterministic evaluator produces PASS, FAIL, ERROR, UNSUPPORTED and NOT_RUN. A timeout configuration is not a measured expiry, and a restart is not original-instance health.
4. Replay an identified correction
Saved inputs retain original manifests, fixture controls and hashes. A candidate must be operator-approved and explicitly selected. Changing fixture mode is an incomparable contrast, not a source-fix proof.
Explore actual recorded cases →