Eleven wrong grades found in the first four hundred transcripts
A registry digitising forty years of paper transcripts discovered that its real exposure was not unreadable paper - it was confidently misread grades.
St Brannock's University
11
grade disagreements caught
2x
independent reads per cell
400
transcripts in the pilot
The challenge
Degree transcripts from the 1980s onward existed only as degraded photocopies, many with toner dropout across the grade column. A previous OCR pilot had returned a full set of grades with uniform confidence - and no way to tell which were guesses.
The approach
- Transcripts taken in as single PDFs so the page run stayed intact through the whole console.
- Bleed-through from double-sided pages removed before any capture ran.
- Every grade cell read by two independent passes, with disagreements surfaced rather than resolved.
- Institutional seals and registrar signatures held and composited back from the original.
Two independent reads found eleven grade disagreements in the first four hundred sheets. All eleven were real.
The results
- Eleven genuine grade disagreements identified and corrected against the paper register during the pilot.
- Captured grades exported as CSV with withheld values in a separate column for reconciliation.
- Restoration and verification folded into one pass, removing a manual checking stage entirely.
This is an illustrative scenario rather than a named customer. It is written to show what a deployment of this shape involves, and the figures describe the pipeline’s behaviour rather than a specific organisation’s results.