---
name: a-watchers-exit-0-with-an-absent-exit-file-is-a-non-result
description: "A wrapper/watcher reporting exit 0 with an EMPTY diagnostics section reads as a perfect green — check the gated command's own exit FILE exists and its raw reached real work; absent exit + truncated raw = nothing ran."
metadata:
  type: feedback
---

Observed 2026-09-08 (hertz, v0.68.0 r2 lane work, reported to me because I keep the
greenlit-form record). A watcher on a killed clippy chain completed and reported
**exit 0** with an empty "first diagnostics, if any" section — the exact shape of a
clean pass. It was not a result:

- the 0 was the **watcher's** exit, not clippy's;
- `clippy.exit` was **ABSENT**;
- `clippy.raw` stopped 11,081 bytes in, mid `Checking anstyle-wincon` — the chain died
  inside DEPENDENCY compilation and never linted a workspace crate.

An empty diagnostics section reads as "no problems found" when it actually means
"nothing ran". That inversion is what gets such an artifact filed as a pass by
whoever reads it a day later.

**Check, in this order, before recording any gate leg as green:**
1. the gated command's own **exit file EXISTS** (an absent exit file is a non-result,
   never a pass — report it as ABSENT and claim nothing);
2. the raw reached **real work** — for clippy/nextest, a workspace crate, not just
   third-party `Checking` lines; compare the raw's size/tail against a known-good run;
3. the reported exit belongs to the **gated command**, not to a wrapper, watcher,
   harness notification, or a pipe's tail.

Sharper face of the standing rule "a verdict comes from the exit file's content, never
a harness completion status" — this is the case where the exit file is MISSING and a
wrapper's own 0 silently stands in for it. Kin: [[never-send-a-claim-composed-before-its-check-ran]],
[[a-fail-prefixed-grep-drops-fail-plus-leak-rows]].