A recorded run

Iris with missing

Asked: Predict variety (one of 3 kinds) from 4 columns of 150 rows.
What happened: 21 ideas tried, 3 kept. It stopped on its own when progress levelled off.
Result: 0.1067Log-loss on data it never saw0.0221 worse than its own estimate, within the normal wobble, so it didn't fool itself.

Try your own data

Recorded without an LLM: ideas come from a built-in search.

Scroll to follow the climb, idea by idea

For the curious

The full record

Every number the run recorded, exactly as measured.