Datasheet · A prompt line
Check examples by hand
Before answering, recompute every given example with your exact formula; if any example differs, your formula is wrong, so rederive it from the stated rules rather than recalling a standard formula.
Designed by OPSIN: it read 24 misses of the unchanged agent on Coding, Haiku 5.5 with thinking off, found 8 of them failed one way (wrong swap formula or fee scaling), and wrote this line aimed at it. Why it fails
Did it fix what it was aimed at? "Check examples by hand" was aimed at wrong swap formula or fee scaling: on the same 96 tasks, 2 such misses with it against 3 without, so it cut the misses it was aimed at. Elsewhere: floored fee before swap 3 against 4; wrong fixed-point scaling in pnl 2 against 3. OPSIN sorted the misses blind to which agent made them; the signed verdict is the result.
Helps on Coding, Haiku 5.5 with thinking off (+3.2 points).
Coding, Haiku 5.5 with thinking off
Pooled across 1 notebook: +3.2 points, 95% interval +1.0 to +5.5; measured once.
- EXCITESNotebook #085+3.2 points (+0.1 to +6.4), 432 pairs, cost ×1.41signed verdict 519
Where it acts
Its effect sits in Parsing durations (+14.1); no other family moved clearly.
- Parsing durations+14.1+2.6 to +29.539 tasks
The other 10 families, none moved clearly
- AMM swap output+10.3−11.5 to +32.039 tasks
- Moving averages+7.7−10.3 to +24.439 tasks
- Trade profit+3.8−15.0 to +20.040 tasks
- Business days+0.0+0.0 to +0.040 tasks
- Fees in whole cents+0.0+0.0 to +0.039 tasks
- Merging intervals+0.0+0.0 to +0.040 tasks
- Run-length encoding+0.0+0.0 to +0.039 tasks
- Comparing versions+0.0+0.0 to +0.039 tasks
- URL slugs+0.0+0.0 to +0.039 tasks
- Top k by score+0.0+0.0 to +0.039 tasks
Read from 432 paired tasks in 1 notebook, each task against its own notebook’s two controls, with every interval widened for the 11 families read. No notebook preregisters a family: this says where to look, and a notebook says whether it holds. The runs behind it are on each notebook’s page.
Put this datasheet in a README
The badge shows what the signed record says about this part, in the setting measured most, and links here. It follows the record as new notebooks publish. Add ?suite= and a suite’s id to show another setting.
[](https://opsinlab.io/parts/6d9e9ff86c1c3bdb)Measure it in a setting it has not been measured in: propose it. As JSON for an agent: /v1/parts/6d9e9ff86c1c3bdb. Every part: the atlas.