GuidesCompare models
Record which model gave the best answer
After a compare run, tell Keenoble which answer won. That judgement is measurement data — it is what re-ranks the library.
Draft — not yet walked in the live product
All 4 steps below were written by reading the Keenoble source rather than by performing them, and each unwalked one is tagged [UNVERIFIED]. Reading a component tells you what a control is labelled; it does not prove the control is reachable, that the order is right, or that the result is what the text says. Until a reviewer has walked every step, this page is excluded from the sitemap, marked noindex, and publishes no structured data.
Before you start
- A finished compare run — every column either done or failed.
Steps
Look below the columns for the row that reads “Which answer was best?”. [Unverified]
It only appears once every column has finished.
Click the button with the name of the model whose answer you preferred. [Unverified]
If two were equally good, click “Tie” instead. If you would rather not judge, click “Skip”. [Unverified]
Skip is recorded as a skip. It is not the same as not answering, and it is not treated as a vote for anything.
The row is replaced by “Logged — thanks.” [Unverified]
If saving fails you get “Couldn't save — try again.” rather than a silent success.
Why we ask
Keen Labs ranks models on evidence rather than brand, and a human preference on a real prompt is evidence of a kind an automatic grader cannot produce. Your picks feed the loop that decides where prompts get routed.
What we publish from that work is the method and the findings — see the research page on this site. What we do not publish is the current routing configuration.
Next
Having problems with the platform? Contact us at support@keenlabs.pro. If a step here does not match what Keenoble does, the step is the bug — send it to the same address. All the guides are at /guides; shorter answers are on the FAQ.