GPT answered Eligible; the verified outcome was Review

Verified answer

△Review Documents conflict on maturity; the conflict must be resolved before a conclusion.

GPT answered

✕Eligible All deterministic characteristics satisfied.

Decisive issue

Deliverability

Evidence for the label

DOC-1#S2DOC-1#S3DOC-1#S4DOC-1#S5DOC-1#S6DOC-1#S8DOC-2#S3

GPT cited

DOC-1#S1DOC-2#S1DOC-9#S91 non-existentevidence score 0%

• Hallucinated evidence. 1 cited section id(s) do not exist in the case record.
• Retrieval failure. None of the cited sections is among those that establish the answer; the relevant evidence was not retrieved.
• Failed to notice contradiction. Documents conflict on a material fact; the answer did not flag the conflict.
• Overconfident. Confidence 0.92 on an answer the record cannot support.

Classifications are produced by the failure analyzer against the verified label and the evidence map; the model never grades itself. Failure taxonomy →

• Mastermind✓Correctanswered Review
• Claude✓Correctanswered Review
• Gemini✕Incorrectanswered Eligibleits failure page →

Open the full Case Replay with the documents →

• #57 Brightwater Energy Corp 2024Wrong final answer
• #67 Republic of Sanderia 2025Hallucinated evidence
• #76 Meridian Telecom S.A. 2026Hallucinated evidence
• #135 Solvane Airlines S.A. 2027Hallucinated evidence
• #195 Brightwater Energy Corp 2028 (B)Hallucinated evidence
• #214 Meridian Telecom S.A. 2025 (B)Hallucinated evidence

All GPT deliverability failures →

This page records GPT’s answer from its latest official run on the frozen benchmark v0.1. If the model is re-run and the answer changes, the page shows the new result and the model page keeps the full run history. Corrections policy →