Washout Failure¶
Core Idea¶
Washout failure is the design defect of underestimating the dissipation time between successive conditions on the same unit, so that an induced state from the first condition contaminates the measurement of the second. The estimate becomes a mixture — the new effect plus the still-decaying tail of the old — because the unit was not returned to a common baseline first.
How would you explain it like I'm…
The Spicy Cracker Trick
If you eat a super spicy chip and then right away taste a plain cracker, the cracker tastes spicy too — but only because the burn from the chip hasn't gone away yet. You have to wait for your mouth to cool down before the cracker can taste like itself. Not waiting long enough fools you about how the cracker really tastes.
Not Waiting Long Enough
Washout failure happens when you test the same thing under one condition and then another, but you don't leave enough of a gap in between. The first condition leaves behind some leftover state that slowly fades, and if you measure the second condition too soon, that leftover is still there. So your reading is really a mix: the second condition's true effect plus the fading tail of the first. The fix is a 'washout' gap long enough for the leftover to disappear — and how long that takes depends on how fast the leftover actually fades, not on picking a tidy round number.
Leftover State Contamination
Washout failure is the design defect where, when the same unit is observed under successive conditions, the second observation is contaminated by residual state from the first because no gap long and clean enough to dissipate that state was built in. The structure is fixed: a unit goes through two or more conditions in sequence; the first induces a state that persists past its cause and decays on some characteristic timescale; a measurement of the second condition is taken before that decay is effectively complete; so the estimate is a mixture of the second condition's true effect plus the residual of the first. Clean sequential measurement requires a latent equivalence — the unit must be in the same state at the start of the second condition as at the start of the first — and the washout interval is the operational proxy for that equivalence. The interval must be set to several decay constants of the carryover process, not a convenient round number, because contamination falls only as fast as the induced state decays. The hidden assumption the pattern exposes is that the unit returns to a common baseline between trials — and washout failure is what happens when that is asserted rather than secured.
Washout failure is the structural pattern in which, when the same unit is observed under successive conditions, the second observation is contaminated by residual state from the first because no gap long and clean enough to dissipate that state was built into the design. It is the design defect of underestimating the dissipation time, so that what is read as an effect of the new condition is partly the still-decaying tail of the old one. The structure has a fixed shape: a unit is subjected to two or more conditions in sequence; the first induces a state that persists past the condition that caused it, decaying on a characteristic timescale; a measurement of the second condition is taken before that decay is effectively complete; the resulting estimate is therefore a mixture — the second condition's true effect plus the residual of the first. The essential commitment is that clean sequential measurement requires a latent equivalence: the unit must be in the same state at the start of the second condition as it was at the start of the first, and the washout interval is the operational proxy for that equivalence. The interval must be set to several decay constants of the carryover process, not to a convenient round number, because the contamination falls only as fast as the induced state itself decays. The hidden assumption the pattern surfaces is that the unit returns to a common baseline between trials; washout failure is what happens when that assumption is asserted rather than secured.
Broad Use¶
- Clinical trials: a crossover design with too short a drug washout biases the second arm's effect estimate with residual pharmacology.
- Education research: testing a curriculum immediately after a prior intervention conflates the new pedagogy with carryover learning.
- A/B testing: re-exposing the same users to a new variant before adaptation to the old has decayed inflates or deflates the lift estimate.
- Policy evaluation: assessing a new regulation while the prior regime's behavior is still adjusting attributes those adjustment costs to the new policy.
- Sensors and metrology: an instrument that has not returned to baseline between samples reports systematically biased successor readings.
- Neuroscience and psychophysics: stimuli presented before adaptation to a prior stimulus has decayed measure against a moving baseline.
Clarity¶
Surfaces the hidden assumption in any sequential measurement — that the unit returns to a common baseline — and attaches a quantity to it, letting an analyst attribute a biased second-period estimate to insufficient settling rather than to a real effect.
Manages Complexity¶
Collapses "is the second-period effect real?" into two answerable questions: what is the carryover half-life, and is the inter-condition gap many half-lives long?
Abstract Reasoning¶
Reframes the estimate as a unit-level counterfactual (same state at each condition's start) and yields a pre-data prediction: bias is a known function of the ratio of the gap to the half-life, and the fix is temporal, not analytic.
Knowledge Transfer¶
- Pharmacology to A/B testing: an analyst who has set a drug washout recognises the same requirement in a between-variant gap matched to the adaptation timescale.
- Across substrates: "pilot to estimate the half-life, set the gap to several decay constants, counterbalance order" is one move in pharmacology, behavioral experiments, and metrology.
- Clinical to engineering: the settling interval set to several time constants is the washout discipline in a non-biological substrate.
Example¶
In a crossover trial a drug with half-life t½ and too short a washout leaves its still-active tail mixed into the read of the next arm; the diagnostic signature is that shortening the gap inflates the estimate while lengthening it converges to a stable value.
Relationships to Other Primes¶
Parents (1) — more general patterns this builds on
- Washout Failure is a kind of Confounding — Washout failure is a specific TEMPORAL variety of confounding: the confounder is the residual state of a prior condition on the same unit, decaying on a half-life, biasing a successor estimate. The file frames it as 'one particular member' of confounding.
Path to root: Washout Failure → Confounding → Bias
Not to Be Confused With¶
- Washout Failure is not Confounding in general because confounding is any third factor distorting a comparison (fixed by randomization or adjustment), whereas this is the temporal variety fixed by waiting for a decaying residual.
- Washout Failure is not Hysteresis because hysteresis is a property of the system (output depends on input path), whereas washout failure is a property of the measurement design that failed to let that history dissipate.
- Washout Failure is not Temporal Decay and Degradation because decay is about a quantity of interest gradually being lost, whereas here the decay is a nuisance corrupting a different quantity, the successor estimate.