performance

Reproduce the baseline, compare equivalent work, and trace the missing time. Acceptance, production and the original plan may differ in more than one way.

Separate startup attestation and model loading from steady-state serving. Compare supported configurations at equal workload and latency objectives, then test whether protected transfers lie on the critical path.

Compare standard-map guarantees with flat-table layouts. Implement bounded probing and choose from semantics and representative measurements.

Connect a performance change to its measured mechanism. Compare fair baselines and calculate how a local speedup affects the whole workload.