Medical records aren't checked in isolation. A prescription in one section can contradict an allergy documented in another. A vital sign can be physiologically impossible. A laterality note in a procedure header can say the opposite of what the body text describes. These inconsistencies create real risk for real patients.
AI systems analyzing clinical documentation need to perform cross-section reasoning, not just read notes line by line. That requires human judgment to identify exactly where language, values, and clinical details don't add up.
The Clinical Documentation Error Detection Quest is now live on Perle Labs! Visit app.perle.xyz to start earning points.
What This Task Is
The Clinical Documentation Error Detection Quest is a set of 40 text highlight tasks. Each prompt presents a realistic clinical note, including admission notes, discharge summaries, operative reports, prescriptions, and progress notes. Each note contains exactly one subtle but clinically significant error.
Your job is to find it and highlight the specific text span containing the mistake. These annotations help train AI systems to detect safety-critical errors in medical documentation before they reach patients.
What You'll Do
Step 1: Read the entire clinical note carefully before highlighting anything.
Step 2: Cross-reference internally — check medications against documented allergies, verify that laterality is consistent throughout, confirm that vital signs and lab values are physiologically plausible, and make sure any interpretations match the actual numbers provided.
Step 3: Identify the one clinically significant error in the note.
Step 4: Highlight only the minimum text needed to identify the mistake — not the full sentence — and submit.
Each note has exactly one primary error. If you spot multiple issues, select the one with the most direct patient safety implication.
Error Categories
- Medication Safety Errors: Drug prescribed despite documented allergy, 10x dosage errors from decimal misplacement, wrong drug class for condition, drug-drug interaction, route of administration error
- Laterality & Site Errors: Procedure header says one side but body text references the opposite side, imaging findings attributed to wrong anatomical location
- Vital Sign & Lab Implausibilities: Values that are physiologically impossible for a conscious/stable patient, decimal errors in lab results, units that don't match the stated value
- Clinical Reasoning Errors: Lab interpretation that contradicts the actual values, diagnosis inconsistent with the presented findings, treatment plan that conflicts with the stated assessment
- Documentation Inconsistencies: Contradictory information within the same note (e.g., age, gender, or history stated differently in two places), monitoring instructions for a drug that doesn't require that type of monitoring
Why It Matters
By contributing to this task, you are:
- Teaching AI to perform cross-section reasoning in clinical documents
- Building enterprise-grade datasets for medical error detection and patient safety systems
- Improving how AI identifies dangerous inconsistencies in real-world documentation
- Contributing to the sovereign data layer that healthcare institutions require
This is expert-level work that AI cannot learn alone. The human layer that makes clinical AI systems safe enough to actually deploy.
Task Details
- Points per question: 150
- Number of tasks: 40
- Total points available: 6,000
- Project eligibility: 2,000 points
Visit the app today to start earning points!
Ready to Get Started?
The Clinical Documentation Error Detection Quest is now live on your Perle Labs dashboard: http://app.perle.xyz
Each annotation you submit helps build auditable AI infrastructure for clinical systems where missing an error isn't an option — not black-box pipelines, but human-verified data with a clear chain of custody.
Stay Connected
Whether you're returning from earlier this season or joining for the first time, Season 1 is open and ready. Follow along with us on X, Discord, and Telegram to stay connected so you don't miss what's coming.
Human-verified. Auditable. Sovereign. Start contributing.
