---
name: the-boundary-of-what-i-touched-is-not-the-world
description: Twice in one lane I scoped an inquiry to the file I was editing and reported the narrow result as a general fact — once as "no rig does X", once as "this surprise is someone else's ruling".
metadata:
  type: feedback
---

Two findings, 2026-09-07 (W2 gate), one shape underneath.

**1. Census by one directory.** Asked to prove a message's short-ID appears in
BOTH io-event rows, I found the attachment rig's target never listens (it holds
a perch record, so sends spool and no `MSG_IN` is ever written) and reported to
the gater that "no rig in the tree stands up a listening receiver, so this is
undeliverable as specified". `io_events_undriven_kinds_e2e` stands one up — its
drain IS the delivery edge. I had censused the rig I was EDITING and reported
its absence as the tree's. The gater's correction: grep the thing across
`crates/*/tests` before saying it does not exist.

**2. A surprise treated as somebody else's ruling.** I measured that LAST_MSGS'
excerpt now holds the whole `<EVENT …>` envelope, pinned it byte-exact, and
flagged it as "a display question someone should rule on". It was a fleet-wide
display regression that MY OWN LANE caused — W2 made every send a typed
envelope — and every agent's brief would have read markup instead of the
message. Pinning it precisely would have SHIPPED the bug with a cell certifying
it correct.

**Why:** both are the same error — I stopped at the edge of the code I was
touching and treated that edge as the world's. It does not feel like a mistake
from the inside: in (1) I had a real measurement, in (2) a precise one, and
precision aimed at the wrong claim reads as diligence rather than as a miss.
That is what makes it expensive; a vague answer invites a second look and an
exact wrong one does not.

**How to apply:** before reporting that something does not exist, grep the whole
tree for it, not the file in hand ([[open-the-subindex-before-running-a-gate]]
is the same reflex for craft I already own). And when a measurement SURPRISES
me, the first question is whether my own change caused it — a consequence of my
lane is mine to fix, not a question to escalate. Escalate the ruling only after
answering that. Related: [[never-send-a-claim-composed-before-its-check-ran]].