APEX-Agents · Investment Banking

world227_tg_02

0/1Fail

APEX-Agents task world227_tg_02 in AI Agents for SEC Disclosure Analysis. Compare dual-harness agent runs across models — rubric criteria, scores, and public traces.

AI Agents for SEC Disclosure AnalysisInvestment Banking World 227Dual harnessGrader: rubric

Task IDtask_9b9afdd77d4240a5bb3c13e83182b91b

WorldInvestment Banking World 227

Expected outputmessage_in_console

Published runs4 models · dual config

Task prompt

What the agent was asked to do

Using Option B in the MFC model, the refinancing in conjunction with a equity private placement (preferred stock), update to upsize the preferred tranche to $1.5bn and for interest to be payment-in-kind (PIK) rather than in cash. In addition, the preferred will now be redeemed at the end of 2030. When the preferred matures, MFC will aim to execute a refinancing for up to 4.5x of gross leverage (gross debt / LTM EBITDA). Use of proceeds from this re-financing will be applied to (a) pay off the preferred; and then (b) fund a shareholder special dividend Calculate the maximum size of the shareholder special dividend in $s to the nearest million. Print the answer 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-5.5showcase	dual	0/1	Fail	Share page Public trace
Gemini 3.1 Pro	dual	0/1	Fail	Share page Public trace
GPT-5.4 mini	dual	1/1	Pass	Share page Public trace
GPT-5.4 nano	dual	1/1	Pass	Share page Public trace

Grading rubric

Criteria and grader verdict (showcase run)

States that the maximum special dividend is $542m
Fail
Evidence: <TEXT_RESPONSE> says "$957 million". Assessment: The criterion requires stating that the maximum special dividend is $542m; the response gives a different amount, so it fails.