Brief — Averages Hide the Knife

A solver can optimise the centre while the tail quietly grows teeth. Takeaway

The average can look safe while the outcomes that matter most — the ugly ones — become more likely.

You can be playing a line that is theoretically sound while accumulating tail risk you do not notice until it bites.

Method

Hold the headline metric constant: EV, win rate or expected return.

Then inspect the distribution, not just the mean:

Compare the median with the mean. Track the tail probability: P(Outcome ≤ threshold) Watch how the tail changes as assumptions shift: opponent response, volatility or execution error.

If the centre remains stable while the tail thickens, the model may be right in the middle and fragile at the edges.

Implication

Optimising the centre is not the same as optimising survivability.

In decision-dense environments, tail risks can compound. One low-probability error may become a cascade when similar situations recur under slightly different conditions.

Apply this

Before trusting any “best line”, set two boundaries:

Downside threshold The outcome you cannot tolerate.

Tail limit The maximum acceptable probability of reaching it.

Then compare the available lines against both boundaries.

Where the average outcomes are close, the line with lower tail risk may be the more robust choice — particularly when conditions are likely to change.