Propose a part

Datasheet · A prompt line

Be concise

Be concise.

Folklore, as cited: a system-prompt staple

Hurts on DeFi maths, Haiku 5.5 with thinking off (−5.1 points); changes nothing that matters on Agent safety, Haiku 5.5 with thinking off; changes nothing that matters on DeFi maths, Haiku 5.5 with thinking on.

DeFi maths, Haiku 5.5 with thinking off

Pooled across 2 notebooks: −5.1 points, 95% interval −7.9 to −2.2; measured once.

  1. UNRESOLVEDNotebook #017−3.9 points (−8.0 to +0.1), 432 pairs, cost ×0.84signed verdict 104
  2. INHIBITSNotebook #018−6.1 points (−10.1 to −2.2), 432 pairs, cost ×0.85signed verdict 110

Agent safety, Haiku 5.5 with thinking off

Pooled across 1 notebook: +1.3 points, 95% interval −0.7 to +3.2; measured once.

  1. INERTNotebook #012+1.3 points (−0.7 to +3.2), 432 pairs, cost ×0.76signed verdict 69

DeFi maths, Haiku 5.5 with thinking on

Pooled across 1 notebook: −0.5 points, 95% interval −3.6 to +2.7; measured once.

  1. INERTNotebook #021−0.5 points (−3.6 to +2.7), 432 pairs, cost ×0.87signed verdict 132

Measure it in a setting it has not been measured in: propose it. As JSON for an agent: /v1/parts/f71fd615fe78a2bc. Every part: the atlas.