Loading
LoadingCheaper Inference / Optimization / Experiments
The experiment keeps assignment and measurement together, so a result can be traced back to the requests that produced it.
Choose the prompt or optimization dimension you want to compare, then describe the alternatives clearly.
Attach an experiment to eligible requests so each variant receives a controlled share of real workload traffic.
Compare billed cost, saved dollars, token usage, request volume, and explicit feedback in one result view.
When the evidence is strong enough, record the winning variant and use it as the next production baseline.
Evidence / 02
Cheaper is useful only when the result still works. Experiment reporting keeps resource use and workspace feedback beside each other instead of declaring a winner from cost alone.