[bagrat]

What

These step outcomes now fail the step, instead of marking the test could_not_run:

• a locator result of stale or provider_error;
• a stale StepExecutionError from the action itself.

As a result, --retries repeats the test. action_uncertain (an action that may have started) and run-wide provider errors behave as before.

Why

--retries only repeats failed tests; could_not_run is final. So one transient hiccup ended a test even with --retries 3:

• a target re-rendered just before the click;
• or a single model call timed out.

In one full run of our 133-test suite, 3 tests ended this way (2 stale clicks, 1 provider error), and all passed on a re-run.

stale is always detected before dispatch, so nothing was clicked and repeating the test is safe. A one-off locator provider_error is scoped to the step; run-wide provider failures still stop the run.

Verification

• Three tests asserted could_not_run for these cases. They now assert failed, and keep their other checks: no re-resolve, no second dispatch, no cache write, no stale replay or evidence frame.
• pnpm typecheck and pnpm lint are clean, and pnpm test passes (1,444).

[coldtea-pr-lens]

• Redraw · drawn at bf1afa1, the branch has moved on

Architecture

Data flow

Tip

The CLI's render reads .github/pr-lens.yml and applies your renames, exclusions and lane pins at draw time

🪧 More tips

• Run npx skills add coldteadotai/pr-lens, then tell your coding agent: "Diagram the change you just made with PR Lens and attach it to the pull request."
• Run npx @coldtea/pr-lens-cli analyze --base origin/main on a branch, then npx @coldtea/pr-lens-cli render .pr-lens/graph.json. Same lenses, your own model key, before the pull request exists
• Click the link under each diagram to open it on a canvas you can zoom, pan and step through
• The diagrams are links. Click one to open it on the canvas, then press W or click play to walk through the change
• Open a diagram on the canvas, then press W or click play to walk through the change one step at a time
• Set github.comment.collapsed: true in .github/pr-lens.yml to fold the comment behind one View architecture and data flow row. Drawing still runs as before
• Set github.draw: on-demand in .github/pr-lens.yml and PR Lens stops drawing pull requests on its own. Comment @pr-lens draw on a pull request when you want that one drawn
• Add .github/workflows/pr-lens.yml with coldteadotai/pr-lens/packages/action@v0 and your model provider's key as its api-key to run PR Lens from your own CI. Any /chat/completions endpoint works
• After a push, tick Redraw to draw the latest commit
• Switch GitHub to dark mode and the diagrams follow. The moving dots are this pull request's data in motion

_{Thanks for using PR Lens! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out on X, Bluesky, Mastodon, Reddit or LinkedIn}

[devnacho]

Thanks a lot for this, @bagrat. The diagnosis was right on: a transient stale target ending the test as could_not_run meant --retries never got a chance.

We landed #169 the next day from the same field report (the 133-test suite). It already covers most of this PR:

• a stale locator result now fails the step (FAILED_LOCATOR_REASONS in sentence-execution.ts);
• a stale action refusal now fails the step (FAILED_ACTION_CODES);
• the action-retry and report tests you updated were switched to expect failed there too.

The one part #169 didn't cover is the locator's provider_error from a single failed model call. I've taken that into #171, with your provider-errors test change, the step's error checked too, a readable message for the step, and you as co-author.

Since this PR now conflicts with main and the rest is already there, I'm closing it in favour of #169 + #171. Thanks again!