Choose Winner
Rank prototype versions by evaluation scores, cost, and latency, then select and promote the best-performing version to production.
What it is
Choosing the winner means picking the best-performing prototype version from your runs so you can promote it to production. In the Prototype dashboard you compare versions by evaluation scores, cost, and latency, set how much each metric matters (e.g. with sliders from 0 to 10), and let the platform rank them. The version with the highest overall score becomes the winner you can then promote, so you move from prototype to production based on data instead of guesswork.
Use cases
- Data-driven promotion — Promote the version that scores best on your chosen metrics instead of shipping by feel.
- Tuning metric importance — Weight evals, cost, and latency differently (e.g. prioritize safety evals over cost).
- Comparing multiple versions — Rank several prototype versions at once and see which one wins under your criteria.
- Production handoff — Select a winner and promote that configuration to production with minimal friction.
How to
Open the Prototype dashboard
Go to the Prototype dashboard and open the project or experiment you want to compare.
Start Choose Winner
Click the Choose Winner button to open the comparison and ranking flow.
Set metric importance
Adjust the sliders for each metric (e.g. evaluation scores, cost, latency) to indicate how important they are on a scale from 0 (not important) to 10 (very important). Your choices determine how versions are ranked.
Review rankings and select the winner
Based on the weights you set, all prototype versions are ranked. The version with the highest overall score is the winner. Select it to promote that configuration to production.