APEX-Agents · Management Consulting
Task 4
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.
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.
| Model | Harness | Score | Result | Links |
|---|---|---|---|---|
| GPT-5.4 | dual | 3/3 | Pass | Run detailsPublic trace |
| GPT-5.4 mini | dual | 3/3 | Pass | Run detailsPublic trace |
| GPT-5.5 | dual | 3/3 | Pass | Run detailsPublic trace |
| GPT-5.4 nano | dual | 1/3 | Fail | Run detailsPublic trace |
| Gemini 3.1 Pro | dual | 0/3 | Fail | Run detailsPublic trace |
Grading rubric
Rubric criteria
Runs are graded against these criteria. Open a run for model-specific verdicts.
States that the total policy-friction risk for North America is $3.146M
States that the total policy-friction risk for the UK is $3.063M
States that the total policy-friction risk for Europe is $3.011M