Several states are possible
A clear single fill can be read as one answer. Very low ink is labelled blank. Two strong fills become multiple marks, and intermediate or close scores remain ambiguous. These states are preserved in the CSV instead of being hidden behind one final percentage.
Correct from the original
Open the student’s source photo when a result is unclear. Enter question=letter after reviewing it; use ? to keep uncertainty or - for a confirmed blank. Corrections apply to all images in a run, so use a single-image batch for student-specific changes.
The displayed score is normalized local darkness, not a statistically calibrated probability. Printer artifacts, erasures, shadows and page curvature can still mislead optical recognition.
Interpret a concrete set of readings
If question 1 shows A:0.92 and the other options near 0.02, that is strong evidence of one dark fill. If question 2 shows B:0.75 and C:0.71, the tool keeps multiple marks unresolved. If all options are near zero, the result is blank. An erased mark or partial fill can fall between these cases and require inspection of the source photo.
Use the answer key only after confirming the form settings. A detected answer can be read confidently and still be incorrect against the key; those are separate decisions. Confirmed correct totals exclude unresolved questions. A low total with many unresolved rows is therefore incomplete evidence, not automatically a final student score.
Make corrections traceable
After reviewing question 2 and deciding B is the intended answer under your grading policy, enter 2=B and rerun. The ledger marks that row as corrected. Use 2=? when the image cannot support a decision, or 2=- for a reviewed blank. The tool does not choose a classroom policy for erasures, multiple selections or accommodations. Apply those rules yourself and retain the exported state labels alongside the original sheet.