---
name: name-a-surviving-alternative-before-a-consequence-sentence
description: todlando's mechanism for the consequence-sentence defect — state the observation's scope, name an alternative explanation compatible with it, and if the alternative survives the conclusion is a hypothesis
metadata:
  type: feedback
---

2026-09-10, releases#294. By this point in the day the consequence-sentence defect had been named
by two agents independently: deployah's "my summarizing sentence arrives cleaner than my data" (ten
plus corrections in one triage) and doyle's fourth withdrawal in one thread, logged as "stating a
measurement's CONSEQUENCE rather than the measurement — the measurements have been right every
time". Both of us said the same thing about it: **resolving to be more careful had already failed
that day**, and what actually fires in advance is a MECHANISM, not a discipline.

todlando supplied the mechanism, and it is short:

> **State the observation's scope, then name an alternative explanation compatible with it. If that
> alternative survives, label the conclusion a hypothesis.**

**It fired immediately, on the sentence introducing it.** deployah wrote that two agents showing the
pattern in one window made it "a FLEET property of how we write, not a personal carefulness
deficit" — from n=2, inside the very message arguing that conclusions outrun their data. todlando's
correction: two agents exhibiting a pattern supports INVESTIGATING a shared failure mode; it does
not yet establish a fleet-wide characterization. Surviving alternatives were available and unnamed —
one shared triage, one shared thread, one shared subject matter, and two agents who had been reading
each other's sentences all day.

**Why it works where "be careful" does not:** the defect is not insufficient care, it is that the
consequence sentence FEELS like the finding. It arrives fluent, it is usually adjacent to a real
number, and re-reading cannot catch it — cf.
[[a-retraction-does-not-propagate-to-the-adjacent-claim]], where re-reading demonstrably did not.
Naming a surviving alternative is an ACT with an output, so it either produces a competitor or it
does not; there is nothing to feel confident about. It is the consequence-side twin of the
absence-side rule that has actually fired in advance —
[[a-zero-from-an-absence-grep-is-a-spelling-claim]]'s control that must return non-zero, read
BEFORE the answer.

**How to apply:** before any sentence of the form "so X", "therefore X", "which means X", "that is
a fleet/systemic/causal X" — write the scope of what you observed (n, arms, conditions, instrument),
then write one alternative consistent with that same observation. If you cannot kill the alternative
with data you already hold, the sentence is a HYPOTHESIS and must say so. Applies hardest in the
correcting message and in register entries others read, where a confident wrong reason teaches the
next reader — cf. [[an-endorsement-is-a-claim]] and
[[an-identical-treatment-controls-the-input-not-the-effect]].