Inspect the work · real engine run
One day-sheet image, and exactly what the engine did with it.
Honest framing: this sheet is a staged, typeset image of a fictional day, not a photograph of real handwriting. What is real is everything below it: this exact image went through the live extraction engine, and the output is shown unedited. A run on a real practice's messy handwritten sheet is the artifact we publish next, with that practice's permission.
The input
What it extracted, unedited
| Time | Patient | CPT billed | Dx | Min | POS |
|---|---|---|---|---|---|
| 9:00 | M.R. | 90837 | F41.1 | 55 | 11 |
| 10:00 | A.T. | 99214, 90833 | F33.1 | 25 | 10 tele |
| 11:00 | D.K. | 90837 | F90.2 | 60 | 11 |
| 1:30 | J.P. | 90834 | F43.1 | 45 | 11 |
| 2:30 | S.B. | 90834 | F32.9 | left empty, not guessed | left empty, not guessed |
Note the 9:00 row: the sheet shows three codes with one circled, and the engine billed only the circled one. Note the 2:30 row: minutes and place of service were unreadable, so they were left empty, not guessed.
The flags it raised, unedited
- Row 2: Telehealth visit (POS 10) is missing the required 95 modifier.
- Row 5: Place of service is missing.
- Row 5: Documented minutes (~35?) are uncertain and likely below the 38-minute floor for 90834.
- Row 5: F32.9 is an unspecified diagnosis code.
Where it asked instead of guessing
- Row 5, minutes: Value '~35?' is ambiguous and contains non-numeric characters.
- Row 5, pos: Cell is explicitly marked '(blank)'.
The human confirm step
Flagged and uncertain rows never become superbills on their own. They come back to the clinician as questions, the clinician answers or corrects, and only then does a document exist. Nothing is filed, sent, or submitted by the engine, ever.