Win/Loss Intelligence
On comparison prompts, who wins the recommendation and why — with per-rival records.
How this engine reports
- Measured evidence — labelled on every number, never dressed up as more certain than it is.
- Works with pasted answers — no API keys required.
- Reads your archived measurement runs.
- Every rate ships with its sample size and a 95% confidence interval; results below n=3 are flagged low-confidence.