ConeReplay and Braintrust.

This is not a scored bake-off. ConeReplay has not published a head-to-head quality, latency, or token number against Braintrust. If you already log traces in another product, ConeReplay does not replace that log.

The question ConeReplay answers

Would this proposed prompt, model, or agent change have changed the outcomes in traces you already recorded? Cone-bounded replay re-executes the causal cone of the edit and serves the rest from the recording, then reports divergence.

Braintrust is a separate product. Use its own documentation for what it stores and how it scores evals. Use ConeReplay when the decision in front of you is a pre-deploy regression check on your own traces.