Repository navigation
feat(tri): select gates by property; the workflow heuristic lied on first use (#2492) - #2513
Merged
Merged
Conversation
Contributor
Contributor
|
📓 NotebookLM Notebook linked to this PR
This notebook contains session context, decisions, and artifacts for this work. |
This was referenced Aug 23, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
#2512 disclosed that the file filter was narrower than the control search. Measuring the hidden files turned that disclosure into three findings, two of them about this command.
A correction first: the disclosure line was itself off by two
It printed
13 of 28where 15 files match by name —rows.len()is the count after the two control files are excluded. The line added for honesty understated the match by exactly the number of controls, and I repeated its number in a report. It now reads17 gate(s) from 28 *.py … control files excluded.Naming was never the property
It failed as
check-vscheck_(#2509), and again asverify_*/run_*. The property that matters is measurable: a workflow invokes it and it can exit non-zero. Anything that can turn a pipeline red is a gate whether or not its name says so.Selection is now by-name or by-property. The count moved 13 → 17, and the three that surfaced —
fuzz_trainer.py,run_conformance_vvp.py,verify_multitarget.py— run in CI, carry verdicts, and have no control in any form. Invisible for the whole campaign to a command whose output looks exhaustive.And the workflow heuristic lied on its first real use
#2512 added it with the explicit warning that upgrading "no control" to "controlled" is the one error direction that hurts here. It then did exactly that, for all three, on the strength of the word
mustsitting in a prose comment 760 lines from the call.Tightened two ways:
fixture,expect_,planted.mustandbrokenare dropped: both are ordinary English in a workflow comment.Both directions are asserted: a fixture path beside the call is found; prose 60 lines away is not.
The cross-check worth keeping
Pointed at the second repository, the tightened version independently reproduced the by-hand measurement from #2511 — the same three gates controlled (
check_build_paths,conformance_check,signal_health), by the same evidence. A heuristic agreeing with a careful manual pass on a corpus it was not tuned against is the closest thing to a control this kind of search can have.15 unit tests green.
Refs #2492