• Product
  • Business Solution

Three jobs, three different ways a file can be wrong.

A missing plate is the easy one. These are the errors that survive a careful person reading the same files twice.

A declared unit with no plate

The varnish plate that went missing.

A seven-unit label went to plate-making as six. Nothing in the files looked wrong. The count was wrong.

Job
JOB-1182
Label
100 ml, 114 × 76 mm
Process
Flexo on chromo
Colour line
6 COL + VARNISH
Customer
a pharma label customer

The job sheet declared 7 units

Six inks and one special.

  1. 01Yellow
  2. 02Magenta
  3. 03Cyan
  4. 04Black
  5. 05Gold
  6. 06P 7483 C
  7. 07Varnishno plate

The separation file held 6 plates

One short. The varnish had no plate.

  1. p1
  2. p2
  3. p3
  4. p4
  5. p5
  6. p6
  7. —

A run printed from that file would have carried no varnish — on a pharma label, where the coating is part of the specification, not a finish.

What the tool reporteddeterministic

Declared 7 units (6 COL + VARNISH), found 6 separation pages. 1 unit(s) missing. Special units declared: Varnish.

Expected
7 units (6 COL + VARNISH)
Found
6 separation pages
Where
Job sheet colour line vs separation PDF page count
  • Plate count

A value that disagrees with itself

Two checks, two opinions, one batch number.

The same field was read in two places and answered differently. Nothing in the report said which reading to believe — so the report itself was the problem.

Field
Batch number, mfg date
Read from
Artwork, composite, each plate
Certainty
Deterministic
Customer
a personal-care label customer

What the old report said

Two rows, formed independently, contradicting each other.

  • Composite row: printed batch details matched.
  • Black plate row: batch number and manufacturing dates differed.
  • Nothing decided which of the two was right.

What it says now

The values are read as values, then compared.

  • The batch number and date are pulled as strings from the artwork, the composite and every plate.
  • A single check compares those strings itself, so a mismatch is arithmetic rather than a judgement.
  • It is reported as a certain finding, and only the source that disagrees is marked, line by line.
  • A value printed on the production files but blank on the artwork is the coding window, and is not a finding.

A batch number that differs between the artwork and the printed plate is a labelling error on a regulated product. It is also the kind of difference two people reading two files will not see.

  • Batch number and manufacturing date
  • Plate-by-plate review

A vendor file that lies about its own shape

The cover page that counted as a plate.

Not every separation PDF starts at plate one. One vendor puts a job specification report in front, and every plate-facing check downstream inherits that off-by-one.

File
Separations PDF
Page 1
Job specification report
Effect
Plate count read 6, not 5
Customer
a plate-making vendor

What the file actually contains

A cover, then the real one-ink plates.

  • Page 1 is a full composite plus a header table, titled JOB SPECIFICATION REPORT.
  • The real one-ink plates start at page 2.
  • Counted plainly, the job reads one plate more than it has.

What the tool does about it

Detect the cover, skip it, keep the page numbers honest.

  • A cover is only detected on page 1, and only when it reads like one — the report title, or two or more of ARTWORK SIZE, OPERATOR, FILE NAME and JOB ID together.
  • Plate count, per-plate identity, colour-name matching and the plate-by-plate review all work from the filtered list.
  • Plates keep the real PDF page number a person would open, cover page or not.
  • Verified against the vendor's own file: the plate count reads 5 of 5 against a declared 5 colours, with plates numbered 2 to 6.

An off-by-one here is worse than no check at all: it reports a plate that does not exist and mislabels every plate after it. A tool that gets this wrong teaches people to ignore it.

  • Plate count
  • Per-plate text map
  • Colour name mapping
  • Plate-by-plate review

The tool approves nothing and rejects nothing. It reads the files, reports what it finds, and says how certain it is. Colour accuracy, trap and overprint intent, and anything that needs a person's judgement stay with the person.