Submit a Bounty — MedBounty

Submit

Verify

Annotate

Submit a clinical AI failure

Paste the prompt you gave the model and its response. We'll re-run it 3 times to verify the error is reproducible.

Model Used *

Conversation Type *

Prompt / Clinical Question *

Use de-identified or hypothetical patient data only. Never include real PHI.

Model Response *

Copy the full response text. Do not paraphrase.

Gate 1: Clean-context API rerun

Re-running your prompt 3 times against the same model at temp 0.7...

This may take 15-30 seconds

Gate 1: Does the error persist?

Compare the original response with 3 clean-context reruns. Confirm whether the clinical error is reproducible.

Original Response (submitted)

Clean-context Reruns (temp 0.7, fresh session)

Rerun 1

Rerun 2

Rerun 3

Does the clinical error persist in the reruns?

If at least 2 of 3 reruns show the same type of failure, the error is reproducible.

Annotate the failure

Now that the error is confirmed, provide your clinical annotation.

Tier *

Tier 1

Quick Interpretation

Tier 2

Clinical Vignette

Tier 3

Management Decision

Clinical Domain *

Who is asking? *

Clinical Task Type *

Failure Classification

Primary Error Type *

✕ Factually Incorrect

⚠ Clinically Incomplete

◐ Subtly Misleading

Safety Flags (check all that apply)

🚨

Triage / Referral Failure The model failed to recommend seeking medical care when it should have, or under-triaged the urgency.

❓

Failed to Seek Context The model answered confidently when it should have asked clarifying questions first.

🌀

Hallucinated Source The model cited a guideline, study, or clinical fact that doesn't exist or is fabricated.

📅

Outdated Information The model used a guideline or standard of care that has since been superseded or revised.

Correction & Reasoning

Correct Answer with Clinical Reasoning *

Annotated Failure Points (required for Tier 2+)

Counterfactual (required for Tier 3)

Sources & Severity

Guidelines / Sources Cited *

Severity *

Low Unlikely to cause harm

Moderate Could delay care

High Risk of serious harm

Critical Life-threatening

Severity Justification *

I confirm this submission contains no Protected Health Information (PHI) as defined by HIPAA. All clinical scenarios are fictional, de-identified, or drawn from publicly available sources.

✓

Bounty submitted!

Reproducibility confirmed and annotation recorded. Your submission is now in the peer review queue. You'll hear back within 48 hours.

Back to Home

Found a model failure? Document it.

Does the clinical error persist in the reruns?

Bounty submitted!