Directory · P
New here? Start with the premise →
Proof
2 min read · 442 words
Proof is what the operator requires before allowing the system to update on a question — and the standard is set by the operator, not by some external rule.
The system holds positions. Some are core (the foundational beliefs, the identity-shaping assumptions). Some are peripheral (the casual opinions, the minor preferences). Each position has a threshold for evidence required to change it. Peripheral positions update easily — small evidence shifts the system’s state. Core positions update reluctantly, sometimes never — the threshold is set so high that no amount of evidence reaches it.
The variable threshold is partly functional. The operator who updated their core beliefs at every passing input would be unstable. Some resistance to updating is appropriate. But the threshold is often miscalibrated — set higher for material that contradicts current belief than for material that confirms it, set lower for sources the operator already trusts than for sources they don’t, set differently for evidence that would require effort to act on than for evidence that supports inaction.
The diagnostic that reveals miscalibration: would I require this same standard of proof for the opposite conclusion. The operator who requires extensive evidence before updating away from their current view, but accepted the current view on much weaker evidence originally, has an asymmetric proof standard that protects current belief regardless of where truth lies. This is common. It is also exactly the configuration that prevents accurate updating.
From the chair: notice when proof requirements are protecting belief rather than seeking accuracy. The signs: the goalposts moving when evidence is provided. The disqualification of sources for reasons not applied to confirming sources. The endless demand for more evidence on questions where the current evidence already exceeds what was originally accepted. These are the system defending the position, not assessing it.
The other application: in the operator’s own work, do not require unreasonable proof before acting. The operator who waits for certainty before deciding is waiting for what reality rarely provides. The operator who requires complete information before acting acts late. The reasonable proof standard for action is — given what I currently know, what would the better-informed version of me likely advise. The answer is usually clear enough. The wait for further proof is often the wait for the system to feel comfortable, which conditions don’t reliably produce.
Proof is calibrated by the operator. The calibration matters. Set it too low and the operator updates on noise. Set it too high and the operator never updates. Set it asymmetrically and the operator confirms what they wanted to confirm regardless of what’s true.