---
name: clippy-compiles-it-does-not-run-tests
description: "A lane that changes diagnostic STRINGS needs the owning crates' test leg — clippy proves it compiles, never that no test pinned the old string"
metadata: 
  node_type: memory
  type: feedback
  originSessionId: 849c3ad4-0c50-4342-b12a-70305d90d7ba
  modified: 2026-08-05T04:15:40.631Z
---

`cargo clippy --workspace --all-targets -D warnings` compiles every target, including test targets.
That is why it FEELS like coverage. It never RUNS a test, so it cannot tell you whether an
assertion pinned the string you just changed.

**Measured 2026-08-04, my owlery-noun lane (`0d1f3e4`), gap named by doyle at gate.** I changed
three user-facing strings — two of them DIAGNOSTIC tokens (`PERCH_CENSUS_SKIPPED`, `PERCH-GC:`) —
and reported three green legs: traceable-reqs, xtask check, clippy. All exit 0, all real. None of
them would have gone red if a test asserted on the old text. doyle closed it himself: swept the
old strings tree-wide, confirmed the tokens appear only in their two emit files, and confirmed no
test asserts on them. **The lane was safe — but I did not know that when I reported it, and that
is the defect.** A green whose blind spot you have not measured is not the same as a green.

**How to apply:** when a lane changes a string that anything could match on — a diagnostic token,
an error message, a rendered help line — run the OWNING CRATES' test leg (nextest) before
reporting, not the whole workspace and not clippy standing in for it. The cheap pre-check when a
full leg is too expensive: grep the OLD string tree-wide and confirm the surviving occurrences are
only the emit sites, with real exits and no pipe swallowing the verdict. Either answers "did I
break a pin"; clippy answers a different question.

**Why it generalises past strings:** the failure was choosing legs by habit rather than by what the
CHANGE could break. A noun swap in copy looks inert, so I picked the copy-lane leg set. The
question is never "what does a lane like this usually run" — it is "what could this specific edit
falsify, and which leg would go red if it had". Related: [[verdict-from-exit-not-from-silence]],
[[retraction-sweeps-diagnostic-strings]], [[a-predicate-without-its-tool-is-not-evidence]],
[[announce-token-from-the-failing-frame-not-the-register]].