Expand description
Which tick made already-passing acceptance tests pass: a fact found by running the tests, never a guess from diffs.
The feature branch’s tick commits are searched by bisection — at most five runs, each with the acceptance files overlaid on the commit’s tree in a scratch worktree of its own under the state directory, sandboxed, under the checks’ deadline. A run that builds and fails says “not passing there”; anything that cannot say — a worktree that cannot be made, a run that times out or prints nothing readable — makes the answer “could not be determined”, never a task’s name. For a section added to a story’s file only that section’s tests run. No model call; the verdict is never changed.
Structs§
Enums§
- Probe
- What one run at a commit said.
Functions§
- budget
- The run budget for
nticks:min(5, ceil(log2(n + 1)) + 1). - prune
- Removes the scratch worktrees a crashed search left behind: only those
whose process is gone, each by its own
git worktree remove— never another live search’s, never the repository’s other worktrees. - search
- Searches the feature’s ticks for the first one at which the acceptance tests pass (they pass on the branch’s tip: the red run said so).
- search_
with searchoverticks(oldest first) withprobe(commit, n)telling what the tests did at a commit. Any undetermined probe ends the search with no answer.- section_
command - The command that runs only the tests
addeddeclares:argvwith the runner’s own filter (None: no filter is known for the runner — the search then cannot run only the section).