bench
Results will appear here
when a run is complete.
Run the benchmark, then publish its normalized scorecard to the connected Vercel Blob store.
Evaluation report
Headroom benchmark
The customer outcomeA0
stock vs B Headroom · provider-observed
Task-level readoutpaired
task / attempt cells
| Task | Outcome | Resolution | Input tokens | Cost | Wall time |
|---|
How the cost was measuredshow your work · not needed to read the result
The headline is a measurement, not a model, and this is the arithmetic behind it. Every dollar of the A0 → B difference is assigned to the price tier the provider actually billed it at; a bar to the right saved money, a bar to the left cost money. The tiers must sum to the observed difference exactly, and the residual below is the check.
Treatment telemetryHeadroom-attributed · B arm only
Published historycompatible
runs
| Run | Date | Harnesses | Status | Complete pairs | Artifacts |
|---|