Processing

Processing

Reading imported sheets, and what to do before you trust the results.

The detailed references are docs/scan_workflow.md and docs/calibration_workflow.md.

Calibrate before a real batch

Do not skip this for real scans. A template can be geometrically perfect and still read badly, because how dark a "marked" bubble is depends on your paper, your printer, your candidates' pencils and your scanner.

On the Calibrate stage:

1. Load the template.
2. Add several representative real scans — including a faint one, a heavy one and one with an erased mark, if you have them.
3. Run recognition in diagnostic mode. You see registration, detected versus expected marker positions, the bubble sampling geometry and the per-bubble fill scores.
4. Adjust the thresholds. Recognition updates immediately without repeating registration, so you can see the effect of a change at once.
5. You get an explicit verdict — passed, passed with warnings, needs review, or failed — rather than a confident-looking result from a template that does not actually fit.
6. Save the thresholds back into the template.

Synthetic sheets do not need this: they are rendered from the template, so they fit it by construction. That is also why calibration cannot be validated with them, and why the Scan stage warns when a template has not been calibrated.

Running a batch

On the Scan stage: Process All, or select rows and Process Selected.

Processing runs in the background. The window stays responsive, and the progress panel shows completed and total counts, percentage, elapsed time, a smoothed estimate of the time remaining, live throughput and per-outcome tallies.

How many cores

Application menu → File → Settings → Processing: Automatic, Single core, or Custom with a worker count. One whole sheet goes to each worker.

Results, output filenames and CSV row order are identical on any number of cores — this is tested, not assumed. Use Single core if you want to rule the pool out while diagnosing something.

If a batch is interrupted

Nothing is lost. Every sheet's result is committed as it finishes, so:

• Resume Batch continues from where it stopped, without reprocessing anything already done.
• Retry Failed re-attempts only the failures.

This holds for a cancellation, a crash, a power loss or closing the window. See Recovery After Interrupted Processing.

Reprocessing

Reprocess re-reads sheets that already have results — after changing thresholds, say. The superseded reading is archived before the sheet is reset, so a sheet processed twice keeps both readings on record.

Reading the outcome

The table shows each sheet's original file, roll number, set, status and output filename. Click a row for the corrected sheet and the recognised values.

Statuses and answer symbols are explained in Recognition Symbols. In short: completed needs nothing, warning means look at it, failed means the page could not be read at all — usually registration.

Anything flagged goes to the Resolve stage.

Related

• Scanning
• Review & Resolution
• Recognition Symbols