Pareto frontier

A configuration dominates another when it is at least as good on every axis and strictly better on at least one. With the default two axes, that means:

cost per successful task is lower or equal, and
success rate is higher or equal,

with at least one of them strictly better.

The Pareto frontier is the set of configurations nothing dominates. Each one is a defensible choice: to get more accuracy you must pay more per success, and to pay less you must accept lower accuracy. Everything off the frontier is a configuration where you pay more for the same or worse results, and ParetoOps recommends eliminating it.

With analyze --max-cost <usd>, configurations whose average cost per task is above that ceiling are also marked dominated, whatever their accuracy.

Choosing on the frontier

Sweet spot: the cheapest frontier configuration that meets your accuracy bar (--min-acc).
Knee point: where extra accuracy starts costing disproportionately more. Past it, each percentage point of accuracy costs much more than the one before.

Small samples

A success rate measured on a few tasks is noisy. The report shows a 95% Wilson lower bound next to each success rate: how low the real rate could plausibly be. Ranking by that bound and adding P95 latency as a third axis are Pro features.