Sample deliverable
This is exactly what you receive
Not a spreadsheet dump - a verifiable report with a folio, a confidence interval, chance-corrected agreement and full traceability back to the raw dataset below.
All data on this page is synthetic - no real patients, no real physicians.
The dataset behind it
Every row that produced this report
The report above is one row of a 12-case synthetic dataset - the same delivery format we ship on every real project: independent verdicts, a reconciled label, and an agreement metric per item.
| case_id | data_type | modality | clinical_question | md1 | md2 | md3 | consensus | agreement % | confidence | adjudicated |
|---|---|---|---|---|---|---|---|---|---|---|
| DLX-0001 | imaging | chest_xray | Is there evidence of pneumonia? | positive | positive | positive | positive | 100 | high | no |
| DLX-0002 | imaging | chest_xray | Is there evidence of pneumonia? | positive | negative | positive | positive | 67 | medium | yes |
| DLX-0003 | imaging | mammography | BIRADS category present? | birads_2 | birads_2 | birads_3 | birads_2 | 67 | medium | yes |
| DLX-0004 | imaging | dermoscopy | Is the lesion suspicious for melanoma? | suspicious | suspicious | suspicious | suspicious | 100 | high | no |
| DLX-0005 | nlp | clinical_note | Does the note document an allergy? | yes | yes | yes | yes | 100 | high | no |
| DLX-0006 | nlp | clinical_note | Is the medication dose extractable and correct? | correct | correct | incorrect | correct | 67 | medium | yes |
| DLX-0007 | llm | model_response | Is the AI response clinically accurate? | accurate | inaccurate | inaccurate | inaccurate | 67 | medium | yes |
| DLX-0008 | llm | model_response | Is the AI response clinically accurate? | accurate | accurate | accurate | accurate | 100 | high | no |
| DLX-0009 | llm | response_ranking | Which response is clinically superior (A/B)? | A | A | B | A | 67 | medium | yes |
| DLX-0010 | transcription | ambient_scribe | Does the transcript contain a clinical error? | error | error | error | error | 100 | high | no |
| DLX-0011 | transcription | ambient_scribe | Does the transcript contain a clinical error? | error | no_error | error | error | 67 | medium | yes |
| DLX-0012 | imaging | ct_abdomen | Is free fluid present? | present | present | present | present | 100 | high | no |
| case_id | Stable identifier for each case - the same one shown on the report folio. |
| md1 / md2 / md3 | Each physician independent, blind verdict. |
| consensus | The reconciled verdict - not a blind average of the three. |
| agreement % | How strongly the panel agreed on this item, raw observed rate. |
| confidence | Delivery confidence derived from agreement and adjudication. |
| adjudicated | Whether disagreement was escalated and resolved by a senior reviewer. |
Dataset-level Krippendorff’s α: 0.78 (example). Only verdicts from physicians whose credentials passed human verification, and who consented to their data being used, ever enter a dataset - no anonymous pool, no unverified reviewers.
Next step
Want to see it run on your cases?
Send a sample sprint through the pipeline - you keep the data and the results.