Skip to content

tri pr ready read 'never ran' as 'never failed' — a verdict built from zero observations #2276

Closed

Description

@gHashTag

tri pr ready classified every failing check against a baseline of failures on recent master commits and on the last few merged pull requests. A check absent from that baseline was printed as:

NOT failing on recent master commits or in the last 5 merged PRs

and counted as introduced by the change under review.

That sentence is indistinguishable for two very different states: the check ran there and passed, or the check never ran there at all. emit-bitexact-gate.yml is the second. It has on: pull_request with a paths: filter and no push: trigger, so:

  • master runs of that workflow, all time: 0
  • last 5 merged pull requests that ran it: 0 (none touched the filtered paths)

The verdict was therefore assembled from zero observations in both windows and printed as evidence. It is the same shape as the errors this repository already catalogues: an instrument that cannot see reporting what it cannot see as a clean reading.

Fix: record what was observed, not only what was red, and give the third state its own verdict — NO BASELINE / CANNOT TELL — which blocks the merge while naming the cause as the repository's CI rather than the change.

Separately, the failure it was hiding is real and unrelated to any pull request: 7 specs no longer generate (parse errors in compiler specs) and 138 specs in specs_generate_baseline.txt now generate again. Filed here as context; the drift needs its own decision.

Activity

  1. added a commit that references this issue on Aug 20, 2026
    39f93d1
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions