Propose a part

Datasheet · A prompt line

Verify by inverse

Do long multiplications and divisions in labelled partial products, then check each by multiplying back or an inverse operation; redo any step that fails before continuing.

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 (mental arithmetic drift in long products and quotients), and wrote this line aimed at it. Why it fails

Did it fix what it was aimed at? "Verify by inverse" was aimed at mental arithmetic drift in long products and quotients: on the same 96 tasks, 16 such misses with it against 18 without, so it cut the misses it was aimed at. Elsewhere: early rounding of intermediates 4 against 14; power or percentage arithmetic slip 2 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 (+21.3 points, replicated).

DeFi maths, Haiku 5.5 with thinking off

Pooled across 2 notebooks: +21.3 points, 95% interval +17.6 to +25.0; replicated.

  1. EXCITESNotebook #078+25.0 points (+10.4 to +40.6), 48 pairs, cost ×2.03, stopped early, provisionalsigned verdict 489
  2. EXCITESNotebook #082+20.8 points (+15.6 to +26.2), 432 pairs, cost ×1.99signed verdict 502

Where it acts

It helps on 9 of the 12 families it was read on, most on Average entry price (+55.0), APR to APY (+35.0), Profit after fees (+33.8) and 6 more.

  1. Average entry price+55.0+33.8 to +76.340 tasks
  2. APR to APY+35.0+20.0 to +51.340 tasks
  3. Profit after fees+33.8+10.0 to +55.040 tasks
  4. Liquidation prices+25.0+8.8 to +43.840 tasks
  5. Swaps+25.0+3.8 to +47.540 tasks
  6. Bonding-curve buys+22.5+2.5 to +42.540 tasks
  7. Position size from risk+17.5+2.5 to +32.540 tasks
  8. Two-hop swaps+16.3+2.5 to +33.840 tasks
  9. Fee splits+12.5+2.5 to +26.340 tasks
The other 3 families, none moved clearly
  1. Token decimals+8.8−7.5 to +25.040 tasks
  2. Impermanent loss+5.0+0.0 to +15.040 tasks
  3. Funding payments−1.3−12.5 to +5.040 tasks

Read from 480 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 +21.3 on DeFi maths, held twice
[![OPSIN](https://opsinlab.io/badge/902767b30d875d72.svg)](https://opsinlab.io/parts/902767b30d875d72)

Measure it in a setting it has not been measured in: propose it. As JSON for an agent: /v1/parts/902767b30d875d72. Every part: the atlas.