--- name: a-named-line-is-a-sample-census-the-whole-file-for-that-shape description: hertz 2026-09-08 lane 2 — dispatch named quickstart_e2e:102; a THIRD stale envelope literal sat at :153 in the same file and only the clean battery found it; when a dispatch names a line, grep the whole file (and the sibling files) for the shape before repinning; also: running a whole e2e binary that launches daemons via WMI in parallel manufactures N identical setup reds — single-thread it metadata: type: feedback --- Lane 2 of the #272 respin (2026-09-08): the dispatch named poll_envelope_e2e:85-89 and quickstart_e2e:102. hertz censused every OTHER file for whole-envelope literals, then read only the named regions in the named files — poll_envelope's second literal (lines[1]) was caught because it sat on the same screen; quickstart's third (:153, the backlog drain assert) was found only when the repinned live line passed and the cell died forty lines lower. Same afternoon: running `-E binary(io_events_undriven_kinds_e2e)` made all seven cells red at the shared setup line :87 (`DAEMON_LAUNCH_VIA_WMI ... did not come up within the start window`) because seven WMI daemon launches contended; identical to seven genuine reds, and the HEAVY classification that spaces them (test/rig-advisory-ports-and-heavy-class) is not on main. **Why:** a named line is a SAMPLE of a defect class, not its population; the fix that repins one sample leaves the next occurrence to red one line lower, on the next run, as "the repin broke it". And a parallel run of daemon-spawning cells measures contention, not the cells. **How to apply:** when a dispatch names a line, grep the whole file and the sibling test files for the SHAPE (here: `