Inbound-transfer point-in-time false-positive backtestRisk & Fraud
How each model performed on this task. Every chart states its N, and the reward carries a 95% confidence interval once there is more than one trial to measure.
Core metrics
Check Pass %
Mean share of the task's checks passing, averaged over trials. N = 10 trials per model.
99%
GPT-5.5
99%
Claude Opus 4.8
49%
Gemini 3.1 Pro
34%
Gemini 3.5 Flash
Extended metrics
Median Time
Wall-clock minutes, median across trials. N = 10 trials per model.
6m
Gemini 3.5 Flash
4m
Gemini 3.1 Pro
4m
Claude Opus 4.8
4m
GPT-5.5
Tool Calls
Tool invocations, median across trials — counted from the trajectories. N = 10 trials per model.
29
Gemini 3.5 Flash
19
Gemini 3.1 Pro
17
Claude Opus 4.8
16
GPT-5.5
Verifier × model pass rate
How often each verifier passed, per model. Green is passing, red is failing — a red band across a row is a verifier nothing clears.
| Verifier | GPT-5.5n=10 | Claude Opus 4.8n=10 | Gemini 3.1 Pron=10 | Gemini 3.5 Flashn=10 |
|---|---|---|---|---|
| checks (19) | ||||
| answer_blocked_cohort | 100% | 100% | 100% | 30% |
| answer_current_row_view | 100% | 100% | 10% | 30% |
| answer_excluded_attempt_exposure | 100% | 100% | 100% | 50% |
| answer_gate_margin | 100% | 100% | 60% | 50% |
| answer_lookahead_correction | 90% | 100% | 10% | 20% |
| answer_point_in_time_view | 100% | 100% | 100% | 40% |
| answer_rollout_decision_flip | 100% | 100% | 0% | 30% |
| deliverable_jira_account_matrix | 100% | 100% | 30% | 40% |
| deliverable_jira_ticket_filed | 100% | 100% | 40% | 40% |
| deliverable_jira_ticket_linked_to_eng14 | 100% | 100% | 40% | 40% |
| deliverable_slack_channel_exists | 100% | 100% | 100% | 40% |
| deliverable_slack_message_content | 100% | 90% | 0% | 0% |
| floor_all_connectors | 100% | 100% | 100% | 40% |
| floor_pin_summary | 100% | 100% | 100% | 40% |
| floor_read_both_source_tickets | 100% | 100% | 100% | 100% |
| floor_sql_investigation | 100% | 100% | 100% | 100% |
| floor_write_back | 100% | 100% | 100% | 40% |
| trace_pending_vs_posted_method | 100% | 0% | 100% | 40% |
| trace_point_in_time_method | 100% | 100% | 60% | 20% |
Failure modes on this task
22 tagged verifier failures, by the stage they were assigned to. Higher is worse.
| Failure stage | GPT-5.5 | Claude Opus 4.8 | Gemini 3.1 Pro | Gemini 3.5 Flash |
|---|---|---|---|---|
| Answer1 mode | 1 | 1 | 10 | 10 |