Propose a part

Datasheet · A prompt line

Keep exact fractions

Do not floor or round intermediate values; with BigInt keep fees as exact ratios (multiply numerator and denominator by the fee's scale) and apply one floor or rounding only at the end, as the task states.

Designed by OPSIN: it read 24 misses of the unchanged agent on Coding, Haiku 5.5 with thinking off, found 11 of them failed one way (floored fee before swap), and wrote this line aimed at it. Why it fails

Did it fix what it was aimed at? "Keep exact fractions" was aimed at floored fee before swap: on the same 96 tasks, 0 such misses with it against 5 without, so it cut the misses it was aimed at. Elsewhere: wrong fixed-point scaling in pnl 0 against 3; wrong swap formula or fee scaling 1 against 2. 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 (+8.3 points).

Coding, Haiku 5.5 with thinking off

Pooled across 1 notebook: +8.3 points, 95% interval +4.0 to +12.7; measured once.

  1. EXCITESNotebook #085+8.3 points (+3.1 to +15.1), 96 pairs, cost ×1.18, stopped early, provisionalsigned verdict 520

Where it acts

No single family moved clearly on its own: it acts a little everywhere.

All 11 families, none moved clearly
  1. AMM swap output+55.6+33.3 to +77.89 tasks, too few to read
  2. Trade profit+16.7+0.0 to +33.39 tasks, too few to read
  3. Moving averages+12.5+0.0 to +25.08 tasks, too few to read
  4. Parsing durations+6.3+0.0 to +18.88 tasks, too few to read
  5. Business days+0.0+0.0 to +0.09 tasks, too few to read
  6. Fees in whole cents+0.0+0.0 to +0.09 tasks, too few to read
  7. Merging intervals+0.0+0.0 to +0.09 tasks, too few to read
  8. Run-length encoding+0.0+0.0 to +0.09 tasks, too few to read
  9. Comparing versions+0.0+0.0 to +0.09 tasks, too few to read
  10. URL slugs+0.0+0.0 to +0.08 tasks, too few to read
  11. Top k by score+0.0+0.0 to +0.09 tasks, too few to read

Read from 96 paired tasks in 1 notebook, each task against its own notebook’s two controls, with every interval widened for the 0 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.

OPSIN: helps +8.3 on Coding
[![OPSIN](https://opsinlab.io/badge/6b56c06d5b4daabe.svg)](https://opsinlab.io/parts/6b56c06d5b4daabe)

Measure it in a setting it has not been measured in: propose it. As JSON for an agent: /v1/parts/6b56c06d5b4daabe. Every part: the atlas.