encounter_id stringclasses 6
values | case stringclasses 2
values | run_ordinal int64 1 3 | model stringclasses 1
value | title stringclasses 6
values | note stringclasses 1
value | messages listlengths 21 53 | review dict |
|---|---|---|---|---|---|---|---|
porphyria-batch-022807 | Acute intermittent porphyria | 1 | gpt-6.1-sol | gpt-6.1-sol — Acute intermittent porphyria — run 1 | Recorded clinical actions and responses; action rationales, when present, are explicitly recorded statements, not hidden reasoning. Catalogs and scores are intentionally omitted from this public showcase, including lookup results and scoring details. Selected clinical actions and findings are retained; catalog navigati... | [
{
"role": "environment",
"content": "A 24-year-old female patient presented to the hospital due to constipation and paroxysmal abdominal pain. She has been looking for medical help for several years. At the moment, she is under the control of a psychiatric clinic and pain management clinic due to idiopathic... | {
"review_status": "preliminary_analytical_review_not_clinician_adjudicated",
"diagnosis_outcome": "target_identified",
"recorded_closure": true,
"strengths": [
{
"text": "Diagnostic recognition was grounded in discriminating questions: dark urine and medication exposure were elicited, and the recorde... |
porphyria-batch-022805 | Acute intermittent porphyria | 2 | gpt-6.1-sol | gpt-6.1-sol — Acute intermittent porphyria — run 2 | Recorded clinical actions and responses; action rationales, when present, are explicitly recorded statements, not hidden reasoning. Catalogs and scores are intentionally omitted from this public showcase, including lookup results and scoring details. Selected clinical actions and findings are retained; catalog navigati... | [
{
"role": "environment",
"content": "A 24-year-old female patient presented to the hospital due to constipation and paroxysmal abdominal pain. She has been looking for medical help for several years. At the moment, she is under the control of a psychiatric clinic and pain management clinic due to idiopathic... | {
"review_status": "preliminary_analytical_review_not_clinician_adjudicated",
"diagnosis_outcome": "target_identified",
"recorded_closure": true,
"strengths": [
{
"text": "History established poorly localized pain, dark urine and carbamazepine exposure. Both PBG and ALA were selected before hemin, ret... |
porphyria-batch-022803 | Acute intermittent porphyria | 3 | gpt-6.1-sol | gpt-6.1-sol — Acute intermittent porphyria — run 3 | Recorded clinical actions and responses; action rationales, when present, are explicitly recorded statements, not hidden reasoning. Catalogs and scores are intentionally omitted from this public showcase, including lookup results and scoring details. Selected clinical actions and findings are retained; catalog navigati... | [
{
"role": "environment",
"content": "A 24-year-old female patient presented to the hospital due to constipation and paroxysmal abdominal pain. She has been looking for medical help for several years. At the moment, she is under the control of a psychiatric clinic and pain management clinic due to idiopathic... | {
"review_status": "preliminary_analytical_review_not_clinician_adjudicated",
"diagnosis_outcome": "target_identified",
"recorded_closure": true,
"strengths": [
{
"text": "Discriminating history elicited dark urine and carbamazepine exposure, followed by PBG testing during symptoms and before hemin. P... |
vipoma-batch-022800 | VIPoma / WDHA | 1 | gpt-6.1-sol | gpt-6.1-sol — VIPoma / WDHA — run 1 | Recorded clinical actions and responses; action rationales, when present, are explicitly recorded statements, not hidden reasoning. Catalogs and scores are intentionally omitted from this public showcase, including lookup results and scoring details. Selected clinical actions and findings are retained; catalog navigati... | [
{
"role": "environment",
"content": "A 56-year-old man was brought to the department of general surgery by ambulance after he had lost consciousness on the street. He is visibly confused."
},
{
"role": "clinician",
"content": "Examine: General appearance (inspection)\nRecorded rationale: Urgent ... | {
"review_status": "preliminary_analytical_review_not_clinician_adjudicated",
"diagnosis_outcome": "alternative_only",
"recorded_closure": true,
"strengths": [
{
"text": "The initial collapse/confusion assessment selected general and head examination, glucose, ECG, sodium and head imaging. The visible... |
vipoma-batch-022758 | VIPoma / WDHA | 2 | gpt-6.1-sol | gpt-6.1-sol — VIPoma / WDHA — run 2 | Recorded clinical actions and responses; action rationales, when present, are explicitly recorded statements, not hidden reasoning. Catalogs and scores are intentionally omitted from this public showcase, including lookup results and scoring details. Selected clinical actions and findings are retained; catalog navigati... | [
{
"role": "environment",
"content": "A 56-year-old man was brought to the department of general surgery by ambulance after he had lost consciousness on the street. He is visibly confused."
},
{
"role": "clinician",
"content": "Ask: Do you have abdominal pain?\nRecorded rationale: Assess for abdo... | {
"review_status": "preliminary_analytical_review_not_clinician_adjudicated",
"diagnosis_outcome": "alternative_only",
"recorded_closure": true,
"strengths": [
{
"text": "General appearance, head examination, glucose, sodium, ECG and acid-base testing were selected during the assessment of collapse an... |
vipoma-batch-022754 | VIPoma / WDHA | 3 | gpt-6.1-sol | gpt-6.1-sol — VIPoma / WDHA — run 3 | Recorded clinical actions and responses; action rationales, when present, are explicitly recorded statements, not hidden reasoning. Catalogs and scores are intentionally omitted from this public showcase, including lookup results and scoring details. Selected clinical actions and findings are retained; catalog navigati... | [
{
"role": "environment",
"content": "A 56-year-old man was brought to the department of general surgery by ambulance after he had lost consciousness on the street. He is visibly confused."
},
{
"role": "clinician",
"content": "Ask: Have you had disorders of consciousness?\nRecorded rationale: Cl... | {
"review_status": "preliminary_analytical_review_not_clinician_adjudicated",
"diagnosis_outcome": "alternative_only",
"recorded_closure": true,
"strengths": [
{
"text": "The collapse/confusion assessment selected general and head examination, glucose, ECG, sodium, head imaging and blood-gas testing. ... |
GPT-6.1-Sol - diagnosis versus holistic care in simulated clinical encounters
Dataset description
An annotated showcase of repeated simulated clinical encounters recorded with ClinicalGym. Each row is one encounter; repeated attempts share a case scenario and are not independent cases.
Rows: 6. Unique case scenarios: 2. Recorded model: gpt-6.1-sol.
Acute intermittent porphyria: 3 recorded attempts.
VIPoma / WDHA: 3 recorded attempts.
Why ClinicalGym exists
ClinicalGym teaches agents holistic clinical reasoning and management so they can offer clinicians better advice and a broader patient picture, not replace doctors. This showcase examines diagnosis separately from a complete and appropriate care plan; more drugs or tests do not automatically mean better care.
This review asks whether diagnostic recognition is followed by a coherent whole-patient plan: acute stabilization, discriminating history and examination, underlying cause, complications, medication context, supportive care, monitoring, reassessment and escalation. Plans stated in a rationale are not credited as completed clinical actions.
Inclusion and outcome interpretation
Three attempts per scenario from one batch, all included, newest first within each scenario. No attempts were excluded or replaced. This is an illustrative set, not an unbiased sample or a general accuracy estimate.
Each attempt is a separate simulation session with the same model and request settings. Clinical results are shown only as the model received them. No human comparator is included.
Exact target submitted: 3 encounters.
Alternative diagnosis only: 3 encounters.
No diagnosis submitted: 0 encounters.
These are descriptive counts for two scenarios, not a general diagnostic accuracy estimate. An alternative diagnosis is not equivalent to identifying the case's underlying target. Closure, diagnostic match, treatment execution, and patient recovery are separate outcomes.
Files and schema
encounters.jsonl: one object per encounter; the dataset file selected by this card.transcripts/*.md: chronological messages, numbered M1, M2, … for review references.README.md: this card.
Each row contains encounter_id, case, run_ordinal, model, title, note, messages, and review. Messages contain only role and content. Roles are clinician (the AI agent, not a human doctor), patient (simulated answer), and environment (observations, results, acknowledgements and notices).
The review field is analyst commentary, not model speech or clinical ground truth. It contains review_status, diagnosis_outcome, recorded_closure, and finding arrays strengths, problems, missed_or_unresolved, limitations. Findings include text, evidence_messages (1-based message numbers), source_ids, confidence, has_private_evidence, and private_evidence_only. The last two flag findings that also rely on material not included in this release, such as catalog lookups or the initial prompt.
Load
from datasets import load_dataset
ds = load_dataset("json", data_files="encounters.jsonl", split="train")
messages = ds[0]["messages"]
review = ds[0]["review"]
The local JSON loader names its split train; this is not a recommended training split.
Preparation and intentional omissions
Catalogs and scores are intentionally omitted, including lookup results, scoring breakdowns, and grading internals; an accepted order is a default status, not a clinical verdict, and treatment results are not shown. Treatment review covers the recorded orders only: selection, dose, route, timing, rationale and relevant omissions. It does not assess administration, efficacy or recovery.
Clinical findings, test values, units and reference intervals are shown exactly as the model saw them, at the point it saw them. Values the model never received are withheld. Action rationales are the model's explicitly recorded statements; hidden reasoning is not included.
The source cases are doctor-authored and reviewed by at least two other doctors. This does not clinically validate the model trajectories. The reviews use the cited guidance and message evidence; they are not clinician-adjudicated validation.
Limitations and safety
A small repeated-run showcase cannot establish robustness, generalization to new cases, or clinical safety. Prior exposure of the model to these cases cannot be excluded. Orders and rationales are not evidence of administration, successful treatment, or recovery. Do not use this dataset for patient care: it is not medical advice. References describe context-dependent guidance, not a mandatory checklist of every possible investigation.
Privacy
Identifiers, source paths, grading details and attachment links are not included. Free-text filtering is not certified de-identification.
Medical review references
A1 — Wang B et al. AGA Clinical Practice Update on Diagnosis and Management of Acute Hepatic Porphyrias: Expert Review. Online 2023-01-13, Gastroenterology 164:484–491. DOI 10.1053/j.gastro.2022.11.034. Full PMC.
A2 — Aarsand AK et al. Practical recommendations for biochemical and genetic diagnosis of the porphyrias. Online 2024-06-28, issue March 2025, Liver International 45:e16012. DOI 10.1111/liv.16012. Full PMC.
A3 — Acute Intermittent Porphyria, GeneReviews. Last update 2024-02-08. NCBI Bookshelf.
A4 — Panhematin (hemin) US prescribing information, revised August 2025; DailyMed record updated 2026-02-26. DailyMed.
A5 — Normosang 25 mg/mL SmPC, text revision 2025-04-09, emc update 2025-04-23. UK emc.
A8 — IPNET, Treatment of the acute attack. Page explicitly last updated 2020-12-07 despite a 2026 site footer. Specialist-network page.
G2 — Kardalas E et al. Hypokalemia: a clinical update. Online 2018-03-14, Endocr Connect 7:R135–R146. DOI 10.1530/EC-18-0109. Full PMC.
G3 — Arasaradnam RP et al. BSG guidelines for investigation of chronic diarrhoea in adults, 3rd edition. Online 2018-04-13, Gut 67:1380–1399. DOI 10.1136/gutjnl-2017-315909. Full PMC.
V1 — Hofland J et al. ENETS 2023 guidance paper for functioning pancreatic neuroendocrine tumour syndromes. J Neuroendocrinol August 2023;35:e13318. DOI 10.1111/jne.13318. Full HTML, publisher PDF via Erasmus.
V2 — de Herder WW, Hofland J. Vasoactive Intestinal Peptide-Secreting Tumor (VIPoma), Endotext. Updated 2023-04-05. NCBI Bookshelf.
V21 — Grozinsky-Glasberg S et al. Controversies in NEN: ENETS position statement on nutritional support. Online 2025-06-25, issue December 2025, J Neuroendocrinol 37:e70062. DOI 10.1111/jne.70062. Full PMC.
V8 — Howe JR et al. NANETS consensus on surgical management of pancreatic NETs. Pancreas January 2020;49:1–33. DOI 10.1097/MPA.0000000000001454. Society PDF, PMC.
References have differing access levels and update dates.
- Downloads last month
- 9