IDA fraud resolution code cohort distributionRisk & Fraud
Start with all trials for auto-written insights, or select a single trial to drill into its verifier results, trajectory, and workspace changes.
GPT-5.5
Mean reward
81.7%All trials
Across 10 trials on this task, reward ran from 75% to 100% — 1 of 10 cleared every check.
- 9.8 / 12avg checks passed
- 21avg turns
- 40avg tool calls
- 3mavg time
Summary · GPT-5.5
GPT-5.5 passed 1 of 10 trials. Failures cluster on answer_suspect_fraud_closures (9/10 trials).
Common pattern
answer_suspect_fraud_closures and deliverable_slack_message_content failed in every unsuccessful trial; what else broke varies across the other 1 verifiers. One reproducible defect, with noise around it.
Why they broke
Most of GPT-5.5's failures on this task were classified as Wrong answer — 9 verifiers across 9 of 10 trials.
Per-trial notes
- Trial 01Missed 2 of 12 checks — answer_suspect_fraud_closures and deliverable_slack_message_content.
- Trial 02Missed 2 of 12 checks — answer_suspect_fraud_closures and deliverable_slack_message_content.
- Trial 03Missed 2 of 12 checks — answer_suspect_fraud_closures and deliverable_slack_message_content.
- Trial 04Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
- Trial 05Missed 2 of 12 checks — answer_suspect_fraud_closures and deliverable_slack_message_content.
- Trial 06Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
- Trial 07Missed 2 of 12 checks — answer_suspect_fraud_closures and deliverable_slack_message_content.
- Trial 08Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
- Trial 09Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
- Trial 10Passed every verifier it was graded on.
Claude Opus 4.8
Mean reward
74.2%All trials
Across 10 trials on this task, reward ran from 66.7% to 75% — 0 of 10 cleared every check.
- 8.9 / 12avg checks passed
- 23avg turns
- 37avg tool calls
- 4mavg time
Summary · Claude Opus 4.8
Claude Opus 4.8 passed none of its 10 trials. Failures cluster on answer_suspect_fraud_closures (10/10 trials).
Common pattern
answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method failed in every unsuccessful trial; what else broke varies across the other 1 verifiers. One reproducible defect, with noise around it.
Why they broke
Most of Claude Opus 4.8's failures on this task were classified as Wrong answer — 10 verifiers across 10 of 10 trials.
Per-trial notes
- Trial 01Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
- Trial 02Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
- Trial 03Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
- Trial 04Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
- Trial 05Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
- Trial 06Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
- Trial 07Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
- Trial 08Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
- Trial 09Missed 4 of 12 checks — answer_code_distribution, answer_suspect_fraud_closures and deliverable_slack_message_content and 1 more.
- Trial 10Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
Gemini 3.1 Pro
Mean reward
73.3%All trials
Across 10 trials on this task, reward ran from 50% to 83.3% — 0 of 10 cleared every check.
- 8.8 / 12avg checks passed
- 85avg turns
- 84avg tool calls
- 19mavg time
Summary · Gemini 3.1 Pro
Gemini 3.1 Pro passed none of its 10 trials. Failures cluster on trace_ida_cohort_method (10/10 trials).
Common pattern
trace_ida_cohort_method failed in every unsuccessful trial; what else broke varies across the other 5 verifiers. One reproducible defect, with noise around it.
Why they broke
Most of Gemini 3.1 Pro's failures on this task were classified as Wrong answer — 9 verifiers across 9 of 10 trials.
Per-trial notes
- Trial 01Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
- Trial 02Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
- Trial 03Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
- Trial 04Missed 2 of 12 checks — deliverable_jira_ticket_filed and trace_ida_cohort_method.
- Trial 05Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
- Trial 06Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
- Trial 07Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
- Trial 08Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
- Trial 09Missed 3 of 12 checks — answer_suspect_fraud_closures, deliverable_slack_message_content and trace_ida_cohort_method.
- Trial 10Missed 6 of 12 checks — answer_code_distribution, answer_confirmed_fraud_closures and answer_suspect_fraud_closures and 3 more.
Gemini 3.5 Flash
Mean reward
20.8%All trials
Across 10 trials on this task, reward ran from 8.3% to 33.3% — 0 of 10 cleared every check.
- 2.5 / 12avg checks passed
- 42avg turns
- 40avg tool calls
- 20mavg time
Summary · Gemini 3.5 Flash
Gemini 3.5 Flash passed none of its 10 trials. Failures cluster on answer_suspect_fraud_closures (10/10 trials).
Common pattern
answer_suspect_fraud_closures, deliverable_jira_ticket_filed and deliverable_slack_channel_exists and 5 more failed in every unsuccessful trial; what else broke varies across the other 3 verifiers. One reproducible defect, with noise around it.
Why they broke
Most of Gemini 3.5 Flash's failures on this task were classified as Wrong answer — 9 verifiers across 9 of 10 trials.
Per-trial notes
- Trial 01Missed 10 of 12 checks — answer_code_distribution, answer_confirmed_fraud_closures and answer_suspect_fraud_closures and 7 more.
- Trial 02Missed 8 of 12 checks — answer_suspect_fraud_closures, deliverable_jira_ticket_filed and deliverable_slack_channel_exists and 5 more.
- Trial 03Missed 11 of 12 checks — answer_code_distribution, answer_confirmed_fraud_closures and answer_suspect_fraud_closures and 8 more.
- Trial 04Missed 9 of 12 checks — answer_code_distribution, answer_suspect_fraud_closures and deliverable_jira_ticket_filed and 6 more.
- Trial 05Missed 8 of 12 checks — answer_suspect_fraud_closures, deliverable_jira_ticket_filed and deliverable_slack_channel_exists and 5 more.
- Trial 06Missed 10 of 12 checks — answer_code_distribution, answer_confirmed_fraud_closures and answer_suspect_fraud_closures and 7 more.
- Trial 07Missed 11 of 12 checks — answer_code_distribution, answer_confirmed_fraud_closures and answer_suspect_fraud_closures and 8 more.
- Trial 08Missed 10 of 12 checks — answer_code_distribution, answer_confirmed_fraud_closures and answer_suspect_fraud_closures and 7 more.
- Trial 09Missed 9 of 12 checks — answer_code_distribution, answer_suspect_fraud_closures and deliverable_jira_ticket_filed and 6 more.
- Trial 10Missed 9 of 12 checks — answer_code_distribution, answer_suspect_fraud_closures and deliverable_jira_ticket_filed and 6 more.