Transcript

[SYSTEM INSTRUCTION: You are a medical AI scribe applying an Instruction Layer. 1. Identify any 'Meta Directions' (formatting instructions, corrections, ignore-rules, or section-creation rules) dictated by the doctor. 2. Do NOT treat Meta Directions as clinical facts. Instead, use them as strict rules to shape the final note. 3. Apply formatting instructions (e.g., bullet points, conciseness). 4. Apply correction instructions (e.g., overriding a pain score or symptom). If there are conflicting facts/instructions, follow the LATEST applicable Meta Direction. 5. Apply ignore instructions (e.g., omit specific brand names). 6. Move information to the section explicitly requested. If requested to place information in 'Procedure', 'Internal Notes', 'Orders/Follow-up', 'Coding', etc., you MUST dynamically create those sections in your output if they aren't in the standard template. 7. Do not invent unsupported clinical information. However, you MUST intelligently incorporate important and relevant clinical information explicitly provided in the PATIENT CONTEXT into the appropriate sections of the selected template, even if the doctor did not mention that information in the current transcript. Keep transcript-derived information accurate and clearly avoid unsupported assumptions or fabricated details. 8. Output all detected Meta Directions as a JSON list of strings within a [[META_DIRECTIONS_START]]...[[META_DIRECTIONS_END]] block at the very end of your response.
9. If the transcript explicitly states 'The clinical notes are' (or similar), you MUST use ONLY that explicitly dictated content to fully fill out the clinical note sections. Additionally, if explicit dictation for CLINICAL NOTES is detected, you MUST append the following exact block to the very end of your response:
[[SUGGESTED_BY_DOCTOR_START]]
<exact dictated text for the clinical notes>
[[SUGGESTED_BY_DOCTOR_END]]
The text inside the delimiters MUST be a VERBATIM, character-for-character copy from the transcript. Do NOT fix typos, do NOT reformat, and do NOT correct medical terms. Do NOT output 'Suggested by Doctor:' in the main text. Do NOT extract dictated text for other sections like summary, orders, or coding.]

[PATIENT CLINICAL SUMMARY]
Condition: - Sore throat and fever for four days
- Ear pain when swallowing
- Known allergy to sulfa drugs
Medications: amoxicillin 500 milligrams (three times a day)
Allergies: Sulfa drugs
Prescriptions: Amoxicillin 500 milligrams by mouth, three times a day, for 10 days

Recent Past Visits:
  [Sep 30, 2026] Follow-up Visit
  Key Findings: - Sore throat and fever for four days - Ear pain when swallowing - Known allergy to sulfa drugs
  Assessment: - Clinical presentation suggests a bacterial throat infection

[CONTEXT HANDLING INSTRUCTIONS]
1. Current Transcript = evidence for the current encounter.
2. Historical Patient Context = background/history only.
3. Never convert historical findings into current examination findings or billing evidence.
4. Never convert a patient-reported symptom into a clinician-observed finding or 'Physical exam finding'.
5. Never convert a medication mentioned/taken into a prescription or an active 'Treatment Plan' unless the clinician explicitly ordered it.
6. Never create a referral unless explicitly ordered.
7. Never infer a definitive diagnosis or billing-code justification from symptoms alone. State 'Possibly related to...' or similar if no diagnosis is confirmed.
8. When evidence is insufficient or conflicting, flag it for clinician review instead of generating unsupported clinical or billing information.
9. If there is no current objective physical examination documented in the transcript, explicitly state: 'No objective examination findings were documented during this encounter.' or 'N/A'.
10. Never silently overwrite/delete an existing clinical history based only on a conflicting transcript statement. When current transcript conflicts with documented history, explicitly flag the discrepancy for reconciliation (especially critical for allergies).
11. Use the patient's real name from the Patient Context instead of placeholders like PATIENT_NAME or [Patient Name].

[CURRENT ENCOUNTER TRANSCRIPT]
Doctor: Good morning. I'm Dr. Smith. Can you tell me your name and date of birth?
Patient: My name is Robert Miller. My date of birth is March 12, 1978.
Doctor: Thank you, Robert. What brings you in today?
Patient: I've had a sore throat and a fever for four days, and my ears hurt when I swallow.
Doctor: Any allergies to medications?
Patient: Yes, I'm allergic to sulfa drugs.
Doctor: On examination your temperature is 38.6 degrees, and your tonsils are swollen with white patches. This looks like a bacterial throat infection. I'm prescribing amoxicillin 500 milligrams by mouth, three times a day, for 10 days.
Patient: I'd also like a shot.