196. Delivery latency targets are budgets of benchmark policy, measured per phase, and never a correctness gate on runners the repository does not own
Context
The owner asked for CI under one minute (2026-10-08), pull requests validated under five minutes (2026-10-10), deploys in minutes rather than hours (2026-10-09), small releases every few tens of minutes (2026-10-13) and releases as fast as possible (2026-10-17). A 2026-10-09 proposal named p95 targets (60 s fast feedback, 2 min merge to UI, 5 min merge to release, 10 min change to verified deploy). The shared pool runs one validate at a time, and wall time on it varies with load the repository does not control. The owner ordered the goals (c208#24, c204#0); no message accepts specific numbers.
Decision
Each delivery phase carries a budget in benchmark policy beside the UI budgets (I2489), with the source of its number (the owner's goal or a measured baseline). A phase over budget is a finding with its trend, shown where the drain and the brief are read. Deterministic local budgets may fail a gate (rule performance-evidence); GitHub wall time is reported, never a red verdict. No required check is removed or weakened to meet a budget.
Alternatives rejected
- Hard SLO gates on GitHub wall time (flaky, trains people to skip)
- Targets in prose without measurement (rule no-counts-in-prose)
- Dropping gates to reach the number
Consequences
I3202, I3203 and I3206 measure and budget the phases; the owner's targets become visible budgets with their gap; achievability on the shared pool stays an open question until measured.
Sources
Promoted on 2026-10-18 from the owner's ChatGPT project (workspace chatgpt-majordomus, issue I3100). The owner has not accepted it; it is proposed for the owner's decision.
chatgpt-majordomus:c130#0chatgpt-majordomus:c130#4chatgpt-majordomus:c195#0chatgpt-majordomus:c204#0chatgpt-majordomus:c204#2chatgpt-majordomus:c208#24chatgpt-majordomus:c228#0chatgpt-majordomus:c244#4chatgpt-majordomus:c244#12chatgpt-majordomus:c291#0