Raycaster / evalsBack to AI Agents for Take-Private Analysis

APEX-Agents · Investment Banking

World224_OS_Task01

0/1Fail

APEX-Agents task World224_OS_Task01 in AI Agents for Take-Private Analysis. Compare dual-harness agent runs across models — rubric criteria, scores, and public traces.

AI Agents for Take-Private AnalysisInvestment Banking World 224Dual harnessGrader: rubric
task_dc0a54118f21451f891311e869b1847c
Investment Banking World 224
message_in_console
4 models · dual config

Task prompt

What the agent was asked to do

Return the Year 1 - Year 5 CAGR for Adjusted EBITDA. Round to two decimal places and display percentages with %. Write your answer straight back. I want you to use this info to update values in the LBO model: 1. Year 2 revenue growth rate increases 50 bps vs original case 2. Cost of revenue for 'subscription' increases by 15% YoY from Year 4 to Year 5

Published trajectories

Agent runs on this task

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

ModelHarnessScoreResultLinks
GPT-5.5showcasedual0/1Fail
Gemini 3.1 Produal0/1Fail
GPT-5.4 minidual0/1Fail
GPT-5.4 nanodual0/1Fail

Grading rubric

Criteria and grader verdict (showcase run)

  1. States that the Year 1 to Year 5 adjusted EBITDA CAGR is 19.12%

    Fail

    Evidence: <TEXT_RESPONSE> is "22.18%". Assessment: The criterion requires stating that the Year 1 to Year 5 adjusted EBITDA CAGR is 19.12%; the response instead states 22.18%, so it does not meet the criterion.