Propose a part

Datasheet · A prompt line

Full-precision intermediates

Never round until the final step: keep every intermediate to at least 8 significant digits, and round only the final result to the requested places.

Designed by OPSIN: it read 24 misses of the unchanged agent on DeFi maths, Haiku 5.5 with thinking off, found 10 of them failed one way (early rounding of intermediates), and wrote this line aimed at it. Why it fails

Did it fix what it was aimed at? "Full-precision intermediates" was aimed at early rounding of intermediates: on the same 96 tasks, 5 such misses with it against 14 without, so it cut the misses it was aimed at. Elsewhere: mental arithmetic drift in long products and quotients 28 against 18; power or percentage arithmetic slip 7 against 7. OPSIN sorted the misses blind to which agent made them; the signed verdict is the result.

Helps on DeFi maths, Haiku 5.5 with thinking off (+3.2 points).

DeFi maths, Haiku 5.5 with thinking off

Pooled across 2 notebooks: +3.2 points, 95% interval +0.3 to +6.0; open.

  1. UNRESOLVEDNotebook #078+3.4 points (−2.4 to +9.3), 432 pairs, cost ×1.21signed verdict 488
  2. UNRESOLVEDNotebook #082+3.0 points (−2.1 to +8.1), 432 pairs, cost ×1.22signed verdict 504

Where it acts

Its effect sits in Average entry price (+21.5) and APR to APY (+16.7); no other family moved clearly.

  1. Average entry price+21.5+0.7 to +41.072 tasks
  2. APR to APY+16.7+0.7 to +31.372 tasks
The other 10 families, none moved clearly
  1. Position size from risk+7.6−6.9 to +22.972 tasks
  2. Two-hop swaps+6.9−2.8 to +19.472 tasks
  3. Token decimals+5.6−4.9 to +16.072 tasks
  4. Impermanent loss+3.5−7.6 to +14.672 tasks
  5. Bonding-curve buys+2.1−9.7 to +15.372 tasks
  6. Liquidation prices+1.4−15.3 to +18.872 tasks
  7. Funding payments−1.4−6.9 to +2.872 tasks
  8. Swaps−1.4−14.6 to +12.572 tasks
  9. Fee splits−7.6−19.4 to +3.572 tasks
  10. Profit after fees−16.7−37.5 to +4.272 tasks

Read from 864 paired tasks in 2 notebooks, each task against its own notebook’s two controls, with every interval widened for the 12 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 +3.2 on DeFi maths
[![OPSIN](https://opsinlab.io/badge/bbe69045e20f4dd4.svg)](https://opsinlab.io/parts/bbe69045e20f4dd4)

Measure it in a setting it has not been measured in: propose it. As JSON for an agent: /v1/parts/bbe69045e20f4dd4. Every part: the atlas.