---
name: proved-the-cell-never-ran-the-lane
description: A green CELL is evidence about that cell only — it says nothing about clippy, fmt, treqs or the rest of the lane that actually admits the change, and "green both platforms" on cell evidence is an overclaim.
metadata:
  type: feedback
---

**A cell result is evidence about a cell. A rider is admitted by a LANE.** Running the one test your
change targets proves the behaviour you intended; it proves nothing about the legs that gate the
change — clippy, fmt, traceability, the other binaries in the same crate.

Measured 2026-09-09, spt-core golden #272 r3 assembly, mine, twice in one lane:
- I reported `88625fa0` as **"GREEN BOTH PLATFORMS (kitsubito 10.183 s, hfenduleam 22.296 s)"**. Both
  numbers were real and both were cell timings. The rider then failed
  `clippy::needless_late_init` at `webserve_attachment_e2e.rs:598` — reproducing identically on BOTH
  OSes at the same file:line.
- `b359e40e` had its shape reviewed and accepted by the gater, and failed
  `clippy::items_after_test_module`: I had appended `converge_within` BELOW `mod tests`.

Neither defect touched the behaviour I had proved, and neither was reachable from the cell I ran.
Under golden's `cargo clippy --workspace --all-targets -- -D warnings` both are hard errors, so both
riders would have red the gate's clippy leg — the leg doing exactly its job.

**Why the overclaim is the interesting part:** "green both platforms" is a sentence about the RIDER.
The evidence underneath it was a sentence about ONE TEST. The two are easy to conflate precisely when
the cell is the thing you have been staring at for hours, and the gap is invisible until a different
leg speaks. Same family as [[a-sound-check-can-carry-an-oversized-sentence]] — a true measurement
carrying a claim wider than its subject.

**How to apply:** before calling a rider green, run the legs that ADMIT it, not only the one that
motivated it — at minimum clippy over the workspace with the gate's own flags. Say which leg produced
each green ("cell x5 green; clippy not yet run") rather than a bare "green". When a peer's gate
ordered specific legs, that list IS the definition of green for that change. And check the toolchain
before blaming drift for a local-only lint: here clippy 0.1.96 `ac68faa20c` matched the runner exactly
and `rust-toolchain.toml` pins it, so the finding was real, not local noise.
Sibling: [[read-a-controls-output-not-just-its-exit-code]] — the wrapper reported exit 0 while the
`.exit` file said 101, and reading the file is what surfaced the first of these two.
