---
name: a-baseline-green-cannot-discriminate-under-a-coin-flip-environment
description: "offering to run a base-sha baseline proves nothing when the environment flips coins — only a RED baseline discriminates, and you cannot order one"
metadata: 
  node_type: memory
  type: feedback
  originSessionId: 880d1285-2246-4429-9b5e-eb4331af141f
  modified: 2026-08-29T18:33:15.327Z
---

doyle ruling, 2026-08-29 (W4/#229 gate). I offered to run the failing tests at the base sha
under full parallel load to upgrade "pre-existing" from an inference to a measurement — framed
as a cycle I would rather spend than hand over an inference dressed as a result. He DECLINED,
with the sharper reason: **"a baseline green is uninformative under a coin-flip environment
(asymmetric instrument)."**

**Why:** when the environment fails randomly, a green baseline is consistent with BOTH
hypotheses — my delta is innocent, or my delta is guilty and the baseline simply won the coin
flip. Only a RED baseline would have said anything, and a red is exactly the outcome you cannot
order. So the run has a ~50% chance of producing a result that means nothing and no chance of
producing a strong negative. The instrument is asymmetric: it can confirm but not refute, which
is the shape of a measurement that feels rigorous and buys nothing.

**How to apply:** before offering to spend a cycle on a discriminating run, ask what EACH
outcome would license. If one outcome is uninformative and you cannot force the other, the run
is theatre — say so and rest on the evidence you have. What actually discriminated on that lane
was cheaper and already in hand: the failing SET MOVED between runs at the same sha (a
deterministic defect does not move), every failure passed ALONE, and the mechanisms were
implausible against the delta. ⚠ The offer itself was not the error — offering to measure rather
than assert is right, and I would make it again; the gap was not pre-registering what each
outcome would license, which is the same discipline as
[[pre-register-what-you-do-in-each-provenance-branch]]. Kin:
[[identical-readings-across-opposite-outcomes-indict-the-meter]] (a control agreeing with the
treatment outranks the finding — here the control could not disagree at all).