---
name: a-retraction-does-not-propagate-to-the-adjacent-claim
description: "Withdrawing a defective inference does NOT repair the sentences built on it — twice in one hour the exact defect just named was re-committed, once inside the correction itself."
metadata: 
  node_type: memory
  type: feedback
  originSessionId: 1f74e1f6-5968-4a83-a94b-b1fd8f62c4ee
  modified: 2026-09-10T08:10:48.117Z
---

**Naming a defect is not removing it. The claims adjacent to a retracted one have to be
RE-DERIVED, not re-read.** Measured twice in ~40 minutes, 2026-09-10, golden 34445961595 triage.

**Instance 1 (mine).** todlando caught me conflating whole-test wall time with an internal poll
budget's margin. I retracted it explicitly and at length. Then, in the SAME message, in the
paragraph headed *"what actually survives"* — the section whose entire purpose was to hold only
what had survived the retraction — I wrote that a cell's 16.8–35.0s spread meant *"the budget has
little headroom against this cell's own variance"* and called it a test-design fact. Same
conflation, forty lines after withdrawing it.

**Instance 2 (doyle, worse class).** He quoted my retraction approvingly, corrected me for the
defect — and committed it nine lines later in his own ruling (*"a 2x slower phase halves panel 1's
effective budget"*, *"16.8–35.0s = a coin against the 30s internal wall"*). He re-committed the
defect he was in the act of correcting. He named the asymmetry himself: I re-committed my own, he
re-committed one he was policing.

**Why it happens:** a retraction feels like it settles the whole topic, so the neighbouring
sentences get RE-READ for consistency rather than RE-DERIVED from the measurements. Re-reading
cannot catch it — the sentence still reads fine; it is the derivation underneath that is void.

**How to apply:** after withdrawing an inference, treat every claim that shares its evidence as
UNPROVEN until re-derived from the raw numbers. Write the surviving set as a bare list of
measurements with no connective prose — connective prose is where the defect regrows. Be most
suspicious of a section titled "what survives"; that heading is where both instances landed.
Related: [[retraction-sweeps-diagnostic-strings]] (the strings half of the same discipline) ·
[[an-endorsement-is-a-claim]] · [[whole-test-wall-time-is-not-an-internal-budget-margin]].