Compare systems fairly instead of chasing one lucky latency
one move: benchmark the workload
Open the code pane immediately, then use the Spec tab, optional design question, tests, and artifact when you want them.
Finish each move before carrying the artifact forward.
A credible benchmark fixes workload, quality target, latency percentile, throughput, and warmup rules so optimizations remain comparable.
Deliverable: State the system invariant in one sentence.Summarize timed request runs into throughput, p95 latency, quality pass rate, warmup exclusion, and an SLO verdict.
Deliverable: Type the deterministic core.Warmup runs cannot inflate results, p95 is deterministic, and a quality failure blocks a performance win.
Deliverable: Record a normal case and a boundary case.Change concurrency, add a slow tail, and lower quality while increasing throughput.
Deliverable: Name the failure signal and its guard.A reproducible benchmark report with workload manifest and SLO decision.
Deliverable: Save the contract, evidence, and handoff note.Warmup runs cannot inflate results, p95 is deterministic, and a quality failure blocks a performance win.
Change concurrency, add a slow tail, and lower quality while increasing throughput.
A reproducible benchmark report with workload manifest and SLO decision.
Deriva-authored extension beyond the supplied PDF. The five moves are Deriva’s implementation contract for this project.