resources

the collapse test (plain english)

sometimes a system says: “nothing unusual happened.” but outcomes worsen across multiple domains, in the same direction, after a trigger. the collapse test asks one question:

who this is for

you can use this method if you are:

• an individual facing coordinated adverse treatment
• a union rep assessing whether a case merits escalation
• a lawyer deciding whether “pattern evidence” exists
• a regulator evaluating systemic failure signals
• an auditor assessing structural drift and governance failure

what you need (minimum inputs)

how to run the test (5 steps)

1. define domains
   list each independent area of system behaviour you can observe.
2. mark outcomes
   for each domain: 0 = neutral/improved, 1 = adverse shift.
3. count adverse domains
   let n be the number of domains that worsened.
4. assume innocence (on purpose)
   give the system every benefit of the doubt: generous probabilities, noise, and correlation.
5. compute plausibility
   if the probability of the observed multi-domain pattern becomes actuarially negligible, the “benign explanation” collapses.

similar mechanics: the coin flip

imagine flipping a fair coin. heads or tails — either outcome is ordinary. but if you flip the coin ten times and get heads every time, something isn't right. you might still call it chance, but the explanation becomes increasingly implausible the more heads you flip.

two views of the same collapse

most people reason about accumulating outcomes additively: one more adverse shift feels like one more comparable “unit” of evidence. That intuition is linear, and it is usually wrong for probability.

in multi-domain alignment, plausibility does not decrease by addition. it decreases by multiplication. Each additional affected dimension scales down what remains, so the innocent explanation collapses exponentially in ordinary (linear) probability terms.

view 1: collapse in ordinary probability (linear scale)

this is what exponential collapse looks like in the scale people are most familiar with. early steps dominate visually; later steps compress toward zero. that compression does not mean later domains “matter less” — it is a consequence of representing a multiplicative process on a linear axis.

view 2: the correct reasoning scale (log probability)

log probability is the correct scale for interpreting multiplicative processes. here, each additional affected dimension produces an equal-sized step because multiplication becomes addition under a logarithm. the boxes are equal in width even though the underlying probability is shrinking rapidly in ordinary terms.

the key point is not that either view is “intuitive”. the key point is that additive human reasoning fails for multiplicative probability. log space is the representation that makes the accumulation law honest.

when many distinct system dimensions move in the same adverse direction, the evidential force comes from dimensionality — not from any single decision.

how to interpret the result

this is a decision-theoretic threshold, not a moral one.

what to do with the result

important boundaries