Every test file has its own budget
A single budget for the whole suite has to be re-measured every time a file is added or removed, and it cannot say which file got slower. A budget per file answers both: the suite’s budget is just the sum, and a new file is measured as part of the runs that happen anyway.
The user’s choice of 2026-10-15: option (B), per-file budgets.
- Each test file has its own measured budget, in CPU work (Q-TX2), and the suite’s budget is their sum.
- A new or changed file is measured within the runs that already happen, so a change in the test population never needs a dedicated burst of runs.
- The ratchet only goes down.
What is missing: not built yet. The web suite’s budgets judge the supported set as a whole (the average CPU per test file, and each file’s share of the worker pool); no file has its own measured budget, and the suite’s budget is not their sum.