Capabilities / D · Medical practice

D.18 Diagnostic reasoning and differential Proposed

Given case material, the agent fills a structured assessment form listing differential diagnoses, each with a status (excluded, to be excluded, considered) and supporting evidence, and gives the order of tests needed to distinguish them. Graded by checkpoints on each listed diagnosis, status and test order; logging of missing information is scored under X.5, not here.

Proposed. This capability is under RFC and does not yet carry levels.

Evidence: planned. Only planned benchmark cards list this capability so far; no frozen benchmark version exists for it.

Domain
D · Medical practice
Evidence form
checkpoint
Veto constraints
—
Depends on
A.1 Medical semantic understandingA.4 Multi-source cross-analysisA.5 Temporal and knowability reasoning
Required by
D.19
Internal task families
—
External anchors
MedQA, PhysicianBench
Version
0.1.0 · 0.1.0

Definition in detail

Safe diagnosis means keeping dangerous alternatives open until they are excluded, and ordering tests to separate them. An explicit differential with status and evidence makes that reasoning checkable.

Inputs a task gives

  • Case material such as history, examination and initial results
  • The list of tests available in the setting
  • The status vocabulary for the assessment form

Artifact required

An assessment form: differentials[] (diagnosis, status as excluded, to be excluded or considered, supporting evidence and refuting evidence as verbatim excerpts) and test_order[] (test, the diagnoses it distinguishes).

How it is graded

Checkpoints cover each reference diagnosis (present or not), its status and the relative order of required tests; where a task cites grading references, there is one checkpoint per reference. Logging of missing information is scored under X.5, not here.

In scope

  • Listing relevant differential diagnoses, including dangerous ones
  • Assigning a status supported by evidence
  • Ordering tests that distinguish the candidates

Out of scope

  • Logging missing information - graded under X.5
  • Urgency and disposition - graded under D.17
  • Choosing and appraising treatment - graded under D.19

Typical failures

Expected failure patterns; not measured rates.

Standards: std-aha-acc-2021-chest-pain · std-wses-2020-appendicitis

Benchmarks

Grading references

No grading reference is mapped to this capability.

Levels by model and harness

No verified results yetLevels would be derived from site-verified result records; methodology 0.1.0 sets no level thresholds, so the level shows not issued.