[lilyshen0722]

Temporary stacked PR to verify that the #2417 path selector skips every test shard on a docs-only diff. Close after recording the run in #2417.

[lilyshen0722]

This probe does not exercise the case its title names, and the case it misses is the risky one.

Titled "docs-only path probe", but its files are backend/__tests__/unit/ciPathProbe.test.js and docs/ci-path-probe.md. Running #2417's own selector over that exact list:

#2418 files        -> {'backend': True, 'frontend': False, 'tools': False, 'chart': True}
a true docs-only   -> {'backend': False, 'frontend': False, 'tools': False, 'chart': False}

The live checks agree — Backend tests (1/3), (2/3), (3/3), Backend typecheck and lint and Chart Lint are all running here, with only CLI, MCP… and Frontend tests skipping. So this is a backend probe. A useful one: it confirms the matrix expands (the literal ${{ matrix.shard }} that shows on #2419 is just how a skipped matrix job renders), and that backend/ correctly pulls in chart.

What neither probe covers is docs-only, where all four suites skip and Test & Coverage must still report success. That is the branch where the new gate could go wrong in the dangerous direction — a required check that passes with nothing run — and it is the one with no live evidence. #2419 covers frontend-only cleanly (frontend runs, the other three skip, Select test paths passes), which matches what I simulated against the gate's jq on #2417.

Dropping the backend file from this PR would make it the docs-only probe it says it is; keeping it and adding a third, docs/-only PR would cover both. Either way the three-case matrix is what makes the experiment conclusive.

PR targets main failing on both probes is expected, not a defect — they are stacked on ci-doctor/tests-split rather than main, which is also the right base for validating that branch's workflow.

[lilyshen0722]

Closing: this was a temporary path probe for #2417, which has merged, so it has no further use. The branch stays in place.