Every proof checked against the spec before the customer signs.

Artwork arrives from the brand or its agency. Your prepress team fits it to the dieline, sets up colors and plates, and sends a proof. The customer marks it up, a second or third round follows, and only a signed approval releases the job to plates and dies. Many corrections in those rounds were checks someone could have made against the paperwork before the first proof went out.

For the test, AI checks copies of past proofs against the spec and copy the customer supplied and writes a difference list for each one: what it compared, what it found and where on the proof. Your reviewer then sets that list beside the corrections your team and the customer actually made.

The dieline has to be the current one. Artwork is often built on a dieline from an earlier revision, or from a sibling size. The check compares the proof's cut and score lines with the dieline revision on the spec, then looks at what sits near them: copy or a barcode crossing a score, art that stops short of the bleed past a cut edge, coating or varnish over a glue area that has to stay clear for the joint to hold. Structural files often start in Esko ArtiosCAD, but the test reads the PDF copies you send.

Barcodes are numbers before they are bars. The GTIN on the proof has to match the one on the spec, and its last digit is a check digit that can be recalculated, so a transposed number shows up. The symbol has to suit the pack: UPC-A or EAN-13 on a retail unit, often ITF-14 on a shipping case, where GS1 guidance calls for bearer bars around the symbol when it prints on corrugated. Bars must be dark on a light background, and red bars fail because most scanners read with red light. The check lists each of these. It doesn't grade print quality, which takes a verifier on the printed piece.

Copy is compared with what the brand sent, never rewritten. The proof's text is read against the customer's copy deck line by line: product name and variant, net quantity in both US and metric units, ingredient and allergen lines on food packs, warnings, country of origin and the recycling label. In a family of flavors or sizes that share one dieline, the common error is a panel carried over from a sibling item, and the check flags it. Whether the copy meets labeling rules stays with the brand owner; the check only shows where the proof differs from what they supplied.

Marks on the pack have to match the paperwork. A case printed with a box maker's certificate has to state the board grade the spec calls for, not the grade from the last job on that die. An FSC label has to carry the right license code, and an FSC or SFI claim has to match the certified board actually bought for the job, which ties the proof to your chain of custody records. A roll label needs the unwind position, core size and copy position the customer's applicator expects; flexible film needs its eye mark where the packaging machine's sensor reads it, and seal areas kept as the spec says.

What the check leaves alone, and your reviewer. Color accuracy, trapping, plate distortion for flexo and screening stay with your prepress operator and the contract proof. The owner names the reviewer, often the prepress manager or the senior customer service rep who runs approvals. They compare each difference list with the corrections that came back in the real rounds: did it catch what the customer caught, what your own team caught, and did it raise anything that wasn't a real problem.

Once your reviewer says yes, each new proof gets a difference list before it goes to the customer, for about 30 days.

Prepress decides what to fix, and people still send every proof and record every approval. The AI never marks a proof approved, and the live month uses the same scoring table.