coordinated-omissionlisted

# Coordinated Omission ## Purpose Establish whether a set of latency numbers is missing the samples that would have carried its tail, and what to do about it. The failure this skill prevents is the SLO signed against a load test whose p99 was structurally incapable of seeing the queue: the generator stopped issuing requests exactly while the system was slow, so the worst moments produced one sample instead of fifty. The mechanism is precise: the measured system delays or suppresses future observations that the target arrival process would have generated. The resulting sample is conditioned on the system being responsive. A 500 ms completed request may still be recorded as 500 ms, while scheduled arrivals during it are absent, delayed, or timed from their eventual send rather than their due time. The distortion can therefore affect sample density, offered load, queue state and the clock used for each value. Counts are necessary evidence, but they do not identify the cause alone. ## Workflow 1. **Establish the generation model explicitly** — closed-loop, open-loop, or semi-open. Whatever it is, it must be a decision, not the tool's default. 2. **Write the stage and clock model.** Count scheduled/offered, generator-admitted, started, server-accepted and every terminal outcome. Preserve scheduled time, actual start, response completion and deadline so generator lag and send-to-completion response time are separable. 3. **Check response coupling.** In a closed popula