Raycaster/ Eval

APEX-Agents · Management Consulting

Task 4

Best published3/3Pass

APEX-Agents task Task 4 in AI Agents for M&A Legal Due Diligence. Compare dual-harness agent runs across models, scores, and public traces.

AI Agents for M&A Legal Due DiligenceManagement Consulting World 129Dual harnessGrader: rubric
task_0a4ad19b76cf4602914e6b8a4f263690
Management Consulting World 129
message_in_console
5 models · dual config

Task prompt

What the agent was asked to do

For 2024 Won/Upsold deals with NCV ≥ 50k, determine the policy-friction risk per deal as NCV × Discount × tier multiplier × tier PFI, where tier PFI is the benchmark mix-weighted sum of Software Customer User Satisfaction Survey Results. After you rank the regions by the total policy-friction risk, please give me the top 3 regions and their respective total policy friction risk (in $M, rounded to three decimal places) in any order. Refer to the following three files: 1) Deal Transactions sheet, 2) the Customer User Satisfaction Survey Results chart in the software pricing trends doc, and 3) the attached policy mix and multiplier charts. Give me your answers as a reply right here.

Published trajectories

Agent runs on this task

Curated dual-harness runs (parsed + original sandbox). Best scored run per model.

ModelHarnessScoreResultLinks
GPT-5.4dual3/3Pass
GPT-5.4 minidual3/3Pass
GPT-5.5dual3/3Pass
GPT-5.4 nanodual1/3Fail
Gemini 3.1 Produal0/3Fail

Grading rubric

Rubric criteria

Runs are graded against these criteria. Open a run for model-specific verdicts.

  1. States that the total policy-friction risk for North America is $3.146M

  2. States that the total policy-friction risk for the UK is $3.063M

  3. States that the total policy-friction risk for Europe is $3.011M