Possible or Definite - Handoff Summarization (Guarded): leaderboard
Possible or Definite? (2026) is a benchmark of 1,200 MIMIC-IV clinical documents with 9,184 annotated diagnostic uncertainty cues, testing whether LLM outputs keep the stated level of uncertainty. This board asks for a clinician handoff summary under a guarded prompt that asks the model to keep hedged statements and add no diagnoses and scores the uncertainty retained rate: among diagnoses the output keeps, the percent whose uncertainty level matches the source. Higher is better.
Metric: Uncertainty retained rate (%) on clinician handoff summarization of the 1,200 benchmark clinical documents (MIMIC-IV notes, 9,184 cue-target annotations over five uncertainty levels) under the guarded (preserve hedged statements, add no diagnoses) prompt: among diagnostic targets the output keeps, the share whose uncertainty level (certain absent, probable, possible, indeterminate, non-asserted) matches the source; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 62.53 | #453 |
| 2 | Claude Haiku 4.5 | 62.51 | #568 |
| 3 | GPT-OSS-120B | 56.18 | #700 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=possible-or-definite-handoff-summarization-guarded · How It Works · Data refreshed daily, snapshot 2026-11-13.