Pharma labeling and artwork proofreading, explained.

Artwork proofreading explained: one approved text, a flat carton and a very small margin for error

One storage change, from a filed variation to a pack on the shelf

1. Regulatory affairs gets the change approvedSeptember 22: the regulator approves a variation, the formal filing that changes a medicine's license. New stability data tightens storage from 30 °C to 25 °C. The approved text is published in 18 languages the next day.
2. The change is traced to every artworkSomeone lists every carton and leaflet carrying the line. In the example, 64 artworks in 22 countries.
3. The artwork house redraws the filesNew versions come back one by one, some of them with new errors the change itself caused.
4. Each artwork is proofed against its approved textEvery character, plus the Braille, barcode and type size. This is where 30 °C on one side panel is caught, or printed.
5. The printer runs, and packaging quality signs the press proofA press proof is the printer's sample from the real press. First batches with the new text follow on December 1. Old stock has to be destroyed and the destruction recorded.

Step 4 is the subject of this guide. The company, product and figures in this example are fictional.

Why a printed carton can pass every line check and still be wrong, and three more surprises

Approved text, artwork, proof and change wave: the four words to learn first

Thirteen terms are defined in the one-page summary.

What a pack has to get right: words, digits, Braille, codes and type size

In shortA carton is five checks stacked on one sheet: the words, the numbers inside the words, the Braille, the codes and the size of the type. Four methods are in use to prove it, and they differ most in which layers they reach and in what they leave behind for a reviewer. Comparing with the previous artwork answers a useful question, and a different one from comparing with the approved text.

People tend to think of proofreading as reading. On a pharmaceutical carton, reading covers one layer. The other four need a different tool each, and an error in any of them can stop a batch.

Four ways to prove a carton, and which layers each one reaches

The first method, read-aloud proofreading, needs no software and is the only one of the first three that notices when a sentence makes no sense. It leaves behind a signature, with no record of what was compared. The second overlays version 5 on version 4 and shows what moved. It is fast and worth keeping, because a correction can push a line onto another panel. But the previous version has no authority: a storage line wrong in versions 3, 4 and 5 gives a clean result every time.

The third aligns the artwork's text with the approved text for that market and reports each inserted, deleted or changed character. The authority is now the right one, so an error inherited from version 1 appears on the first run.

Its weakness is what extraction can read: print-ready files often carry type converted to outlines (letter shapes with no text behind them), or a language block placed as an image. A tool that reads only live text can report a clean comparison on a panel it never read. The fourth adds the layers that are not text at all.

The labeling rules behind a proof, from 21 CFR 211.122 to the 2026 NDC rule and draft Annex 22

In shortNo rule in the US or EU says how to proofread an artwork file. The rules govern what a pack must say, how printed materials are controlled and how signatures work, and the company designs the proof to meet them. Five moved in 2026, among them FDA's National Drug Code (NDC) rule and the labeling guidance from the UK regulator (MHRA). Annex 22, the draft EU good manufacturing practice (GMP) annex on AI, and the EU pharma package are not yet final.

What each text asks of a carton, a leaflet or a signature

The one sentence that names a second reader

A US firm using cut labeling, labels supplied as separate pieces, must pick one of four controls: dedicated lines, 100-percent machine examination, 100-percent visual examination of hand-applied labels, or an automated technique that physically prevents wrong labels from being processed. The third adds a person: "Such examination shall be performed by one person and independently verified by a second person."1 That sentence is about finished labels on a line. A procedure that cites it as authority for a two-person read-aloud of an artwork file cites the wrong thing.

The rule also rewards the machine route. Reconciliation, the count showing that labels issued, used and returned add up, "is waived for cut or roll labeling if a 100-percent examination for correct labeling is performed in accordance with § 211.122(g)(2)", which is machine examination.2 The visual route earns no such waiver.

What the unfinished EU pharma package would change on a pack

Electronic product information (ePI) is still being prepared: the page of the European Medicines Agency (EMA), updated September 16, 2026, lists a draft roadmap, guidance that will apply at go-live and user testing, and gives no go-live date.25

What a proof compares: the approved text, the print-ready file, the item master and your own rules

In shortA proof needs five inputs: the approved text as a named version, the print-ready artwork, the printed-material rules for that market, the product codes from the item master, and your written severity rules. The approved text is the one that usually causes trouble, because it often lives as a document where a check needs versioned statements mapped to panels.

The mock-up a regulator sees, and the pack you print

The Commission's packaging guideline defines a mock-up as "a copy of the flat artwork design in full colour", cut and folded where necessary to show the text in three dimensions, and a specimen as "a sample of the actual printed out" packaging.6

For centrally authorized medicines, those licensed for the whole EU through EMA, the agency "will perform a general check of mock-ups and specimens from the viewpoint of readability", looking at layout, font sizes and differentiation between strengths.6 That is a readability check of representative packs. Every other pack size, language and version is the company's own work and the company's own evidence.

A change to the labeling text can itself be a filing. For a centrally authorized medicine, an Article 61(3) notification covers labeling or leaflet changes not connected with the summary of product characteristics (the prescriber's reference document), and EMA answers within 90 days. Its own list of examples includes "Change in Braille (inclusion/deletion/change)" and the introduction of QR codes.16 Adding a strength to the Braille line is therefore a regulatory act, whatever it looks like on a designer's screen.

Where the 14 mandatory particulars sit on one flat carton

The law lists what a pack must carry, and those required items are called its particulars. This is how the 14 on the example's German carton sit on the flat sheet, with what can go wrong on each panel.

The layout is the fictional Ostravel carton. Real carton layouts differ by product and market, but the list of things that can go wrong on them repeats.

How one storage change becomes 64 artworks, and what that does to a labeling team's calendar

In shortLabeling change management starts with a fan-out: a single approved statement spreads by market, language, pack size and component. Proofed by hand at 16 working hours each, the example's 64 artworks take 1,024 hours inside 27 working days, close to five people doing nothing else. Three clocks sit above the team: the printer's slot, the regulator's clock and the six-month limit on packed stock.

The Ostravel storage change touches 43 cartons and 21 leaflets. Germany alone has three carton sizes and a leaflet. Switzerland has three leaflets, one per national language. A shared Nordic carton carries four languages, and a Belgian one carries three.

The arithmetic of a change wave at 16 hours an artwork

The 16 hours is the fictional company's estimate and the 3 hours is a design target. Neither is a measurement. No public source publishes hours per artwork proof; section 18 shows how to find yours.

Three clocks a late artwork has to beat

Type size, x-height, check digits and Braille cells: each carton check worked with real numbers

In shortEach check on a pack is arithmetic someone can do by hand. A 6.5-point line fails because 1.2 mm of x-height (the height of a lower-case x) is under the 1.4 mm floor. A Data Matrix product code is valid only if its last digit computes. Ostravel 20 mg in Marburg Medium takes about 86.5 mm of a 110 mm panel. A dropped number sign turns "20" into "bj".

Why a line can pass on points and fail on millimeters

The European floor for labels has two forms: "at least 7 points (or of a size where the lower case "x" is at least 1.4 mm in height), leaving a space between lines of at least 3 mm".7 The MHRA repeats it and adds that it applies to small containers and blister foils too.15

One point is 25.4 ÷ 72 = 0.3528 mm, so a 7-point line has an em (the full height of the type) of 2.47 mm. To reach 1.4 mm at exactly 7 points, a typeface's lower case must be at least 1.4 ÷ 2.47 = 57% of the point size. Many common text faces are smaller than that, which is why a specification should say which test it holds the artwork to.

Two details get lost in briefs. The guideline's 9 points for leaflets is "as measured in font 'Times New Roman', not narrowed", so a condensed cut does not qualify; the older 8-point allowance applied to applications until February 1, 2011.7 And its 3 mm between lines is not defined as baseline to baseline or as clear space, so write down which you use and measure it the same way each time.

In the US, FDA's 2022 guidance gives prescription cartons recommendations and no numeric floor. Over-the-counter Drug Facts labeling has a floor: "no smaller than 6-point type", at least 0.5-point leading (the space between lines) and no more than 39 characters per inch.26,21

Two check digits on one German carton

A check digit is the last digit of a code, calculated from the others, so a mistyped code fails the sum.

The company is fictional. These two numbers are the sample PZN and product code GS1 Germany prints in its own guideline.29 A check digit catches most typing errors. It does not tell you the code belongs to this pack, which is why the code is also compared with the item master.

What the Data Matrix must carry, and what a file can never tell you

The unique identifier holds a product code, serial number, batch number and expiry date, with a national number where a member state requires one, and manufacturers "can choose the order of the data elements".9 Serial numbers and batches are printed on the packaging line, so on an artwork file you can only check that their place exists.

The rule on print quality sets a result the printed code has to achieve: the code must stay readable until at least one year after expiry or five years after release for sale, whichever is longer. "The use of a printing quality of 1.5 or higher gives a presumption of conformity."9 In plain words, a code printed at that quality is assumed to meet the rule. For a pack with a 36-month shelf life, that means five years.

A print grade is a measurement on a printed sample with an instrument. It depends on ink, board, press and geometry. A PDF cannot show it. What a file can settle is the contents, the digits, the fields, the human-readable headers, the size and the clear space around it. Any company that writes "grade B or better" in its specification has made a stricter choice of its own, measured at the press, and should say so in those words.

The Braille arithmetic, and why digits are the trap

The MHRA publishes the numbers for Marburg Medium, the recommended Braille cell size: a 1.6 mm dot, "exactly 2.5mm" between dot centers, 6.0 mm between characters, 10.0 mm between lines and 12.0 mm for a word space.8 The guideline asks for uncontracted Braille, with every letter spelled out, and "In case of multilingual packaging, the name in Braille has to be printed in all the different languages concerned".7

Braille has no separate digits. The first ten letters do double duty, and a number sign in front switches them to numbers. Leave out that one cell, and 2 and 0 read as b and j: "Ostravel bj mg". No reader of the printed side would see the difference, since the printed side is correct. Exemptions exist for packs that only professionals handle, with vaccines as the guideline's own example.7,8

The four kinds of artwork error a careful proofreader's eye is built to skip

In shortThe errors that survive to print are small and specific: a number inside correct words, a difference in a layer nobody reads, a measurement, and something that was never supposed to change. Published evidence on second checkers is thin and comes from neighboring tasks. Nobody has published what two readers catch on artwork, so the only number you can defend is one you measure yourself by planting errors.

Three of those four errors sit nowhere near the sentence that changed, where nobody was told to look. That is why a check aimed at the change misses them, and why more training rarely helps. Other common causes include a shared side-panel file used by sibling pack sizes, a translation pushing a mandatory line onto another panel, and a code rebuilt by hand.

What published studies of second checkers can and cannot tell you

We found no study that measures read-aloud proofreading on pharmaceutical artwork. The closest observational work followed 690 double checks of chemotherapy. Half the inconsistencies came from comparing the order with the label, and half from the nurses using their own knowledge. Three in four traced back to inappropriate orders, made much earlier.31 Applied with care to a proofreading desk, it suggests a split: the comparing half is mechanical, and the knowing half is a person noticing that a pack size and a count disagree.

Pack design research points the same way. In one experiment, 84 adults judged from images whether two different packs held the same active substance. Error rates were 41% for younger and 68% for elderly participants, and fell to 8% and 16% with packs redesigned to highlight the substance.32 That is a laboratory study on generic packs, and it shows that meeting a type-size floor does not prove a pack works.

A review of look-alike labeling found 16 studies, 13 in a laboratory; of 11 on Tall Man lettering (capitals inside look-alike names), six found lower error rates and one found higher.33

How large a seeded-error test has to be before a clean result means anything

A company cannot infer its detection rate from the errors it happens to catch, because that sample says nothing about the ones it missed. The fix is a seeded-error test: plant known differences in real retired artworks, run them through your normal process, and count. It also tells you how much a clean result proves. If a method misses none of n planted errors, the true miss rate could still be as high as 1 − 0.05^1/n at 95% confidence.

What FDA recall data says about labeling errors: 135 events in six years, and a share that depends on the definition

In shortWe counted FDA's own enforcement data for report dates 2020 to 2025. Reasons beginning "Labeling:" cover 135 of 1,901 recall events, about 7%, steady at 20 to 26 a year, and 33 of them include a Class I product. Other wordings of the same reason bring it to about 9%. A raw word search reaches 13%, inflated by phrases such as "below label claim". Label mix-ups are the largest group, and we read most of those as line and material failures that a desk proof cannot prevent.

Read that chart as a map of where the chain breaks, and resist reading it as a verdict on proofreaders. A mix-up is usually the right words on the wrong product, which is what dedicated lines and machine examination exist to stop. Lot and expiry errors are variable data added during packing. A desk that reads files has little influence on those 68 events.

It has more influence on declared strength, missing warnings and wrong inserts: about 35 of the 135. Even for those, the public text rarely says whether the artwork was wrong or a correct artwork went onto the wrong run.

One Class I reason from August 2024 reads like a proof that never happened: "Label Error on Declared Strength; Some cartons were incorrectly labeled as 0.125 mg instead of 0.25 mg. The blister strips inside the product carton reflect the correct strength of 0.25 mg."5 The blisters were right and the carton was wrong. These numbers cover the US only. We found no matching public count for the EU, where defect reports are handled by member states.

What inspectors cite on labeling controls, and what a late artwork does to supply

In shortFDA's published inspection observations (Form 483) show that labeling sections draw few citations, 37 of 2,837 drug citations in fiscal 2025, and only one on examining materials before packaging. That low count is no evidence of good artwork: the citations concern the physical control of labels, and few ever look at how a file was proofed. The larger risks are a recall, destroyed stock, and a release blocked by a missed six-month date.

Why few citations should not reassure a labeling lead

The labeling sections cited most in fiscal 2025 are about the physical object: strict control over labeling issued, ten times; label storage access, five. Examination before packaging appears once, and correct labels during finishing operations once.35

An inspector usually meets artwork as a record. So the question an inspection can ask is what evidence exists: which approved version this carton was checked against, who signed, and what the signature meant. A signed checklist saying "proofread by two readers" is an assertion. A report naming the version, the panel and each difference can be reread two years later and still mean something.

Signed electronic records must show "The printed name of the signer", the date and time, and "The meaning (such as review, approval, responsibility, or authorship) associated with the signature".20 The MHRA's data integrity guidance describes the expectation as ALCOA: attributable, legible, contemporaneous (recorded at the time), original and accurate, with "+" adding complete, consistent, enduring and available.36 A file named "final_v3_FINAL.pdf" on a shared drive is hard to defend under any of those words.

What it costs when a wrong artwork is found, by when it is found

We found no public source that prices these steps, so this table gives no figures. Section 18 shows how to cost your own.

Printed components are controlled material. EU GMP says outdated printed material "should be destroyed and this disposal recorded", and suppliers of it "shall be accorded attention similar to that given to starting materials".22 Any significant discrepancy in reconciliation "should be investigated and satisfactorily accounted for before release".22 US rules read the same way.1,2 A wrong carton found after the run therefore costs more than the board and the ink: it costs a run that has to be destroyed, recorded and replaced while the product waits.

For the patient, critical information has to be found and read correctly, and the MHRA's stated purpose is to help users select and use the right medicine and minimize medication errors.15 FDA's guidance lists crowded labels and hard-to-read text among the design problems that lead to errors.21

The 12-digit NDC: the label change every US drug labeler now has to plan

In shortFDA's final rule moves every National Drug Code to 12 digits from March 7, 2033, with a three-year transition. New labels cannot be used before that date, but they can be drawn and proofed now. FDA expects 95% of updates to ride a change a firm was making anyway, which only works if you know which packs have a change coming.

Today's NDC has one of three shapes. Under the rule, "all FDA-assigned NDCs will be required to be 12 digits in length", with a 6-digit labeler code, 4-digit product code and 2-digit package code.14 An existing code converts by adding leading zeros, and the rule treats that as an administrative change, so the product keeps its identity. On one carton it is a one-line edit and a new barcode. Across a portfolio it is a labeling program.

What to do now, when nothing may be printed until 2033

Other changes that arrive as artwork waves

UK-market packs have carried "UK Only" since January 1, 2025, and EU safety features no longer apply in Northern Ireland.13 Medical device makers moved to FDA's Quality Management System Regulation on February 2, 2026, which incorporates ISO 13485:2016.37 A new company address, a new pack size or a change in product text each fans out the same way, and QR codes on packs raise their own notification question.16 Each of them is a list of artworks with a deadline.

What the pharma artwork market looks like, and the numbers no one publishes

In shortWe found no regulator or dataset that publishes the size of the artwork proofreading market, the cost of an artwork error, versions per artwork or a proofreader's catch rate. Vendor figures come without a dataset, so we left them out. The solid numbers are about scale: 271,655 listed US packages, and one change reaching dozens of artworks.

Numbers you may be quoted, and how to treat them

Why more change is coming to the same teams

Portfolio-wide events such as the NDC format are on the calendar.14 Regulators keep revising the texts a carton has to reflect.6,15 EU reform and electronic product information are in preparation, with no date you can plan a pack around yet.23,25 Meanwhile, a labeling change is rarely anyone's main job, and the people who proof it are usually also running the next one.

AI in artwork proofreading: what draft Annex 22, FDA and EMA ask, and where the limits sit

In shortSoftware can read, compare and measure every character without tiring. It cannot judge whether a deviation is acceptable, and a regulator will expect a named person to decide. The most useful text is a draft: EU GMP's Annex 22 says acceptance criteria must be at least as high as the process replaced, so you must first know how good your readers are. Its line on language models is under review.

The draft annex's clauses that reach an artwork desk

The annex addresses models embedded in computerized systems used in manufacturing, so artwork proofreading sits beside its scope. We apply it here by analogy, on purpose.

Other AI and software texts, with their exact reach

What software changes, and what stays with people

No regulator approves proofreading tools. A company validates its own use of one: intended use written down, acceptance criteria set before testing, test data kept independent of anything used to tune the tool, change control, and monitoring in operation.12

Drift, a change in what the tool is given or reads over time, is concrete in this work: a new artwork house, a new language, or a new export setting in a design tool can change what extraction reads. No software can certify your compliance with Part 11, Annex 11 (the EU GMP annex on computerized systems) or GMP. It can only give you controls that support it.

How labeling teams proof artwork today: who holds what, and where a proof's 16 hours go

In shortFive roles hold the chain: the text, the file, the proof, the press and the release. Only the last is written into law. A proof by hand usually takes two readers and a sign-off, tracked in a spreadsheet, and most of its hours go to reading. Teams run it in house, through an outside service, or with software, and most combine them.

Regulatory affairs to Qualified Person: who holds which part of a pack

Apart from the Qualified Person, no European or US rule we found describes these roles. Titles and splits vary from company to company, and a medical device or generics firm may fold several into one person. The artwork house draws the file and should not approve it: approval is a claim about the approved text, which the house does not own.

From approved text to press proof: where the hand-offs happen

A hand proof's 16 hours, step by step

A return restarts the reading, because a corrected file can move a line break, reflow a column or shrink a three-language block, and the new version has to be read in full. The re-proof of the German carton's version 5 compares the same 2,947 characters as the first pass, for one corrected number.

In-house readers, an outside service or artwork proofing software

The German 30-tablet carton: from a wrong temperature on side panel B to a signed proof report

In shortThis is the example from the top of the page, inside the product. Six specialist agents, software workers with one job each, find every affected artwork, read it, compare it and check it. Annika Brandt, the global labeling lead, sees the whole rollout on one strip, opens the German carton on its difference, returns it to the artwork house with the reason already written, and signs the corrected version with the meaning of her signature.

1

All 64 artworks of one storage change on one strip

Annika's proof screen opens on change wave CW-0418. Each tile is one artwork, grouped by market from Germany to Croatia: 52 passed, 4 with differences, 5 being proofed, and 3 awaiting artwork from Halden Pack Design. The first red tile is the German 30-tablet carton.

Above the tiles sits the change itself, not above 30 °C to not above 25 °C, and the date that matters: printers by October 30.

2

"Go to difference" stops on the 30 °C

ART-DE-C30, version 4, lies flat with its panels and flaps. 2,947 characters on six panels have been compared against approved text DE v2026-09. A heat layer marks what differs, and a wipe moves between the artwork and the approved text set in the same place.

The view zooms and stops on side panel B. Approved: "Nicht über 25 °C lagern." Artwork: "Nicht über 30 °C lagern." Critical. The note beside it says the German leaflet, version 6, already says 25 °C and was signed on October 12. Two items were ignored by rule, a line break and a soft hyphen, and each is listed.

3

The rest of the carton passes, check by check

Six of six checks pass on this carton. The Data Matrix product code decodes as 04150123456782 and its check digit computes. The PZN's check digit computes too. The Braille cells read "Ostravel 20 mg" in Marburg Medium. The smallest line is 7.4 points with a 1.5 mm x-height. All 14 mandatory particulars are on the right panels, and the tamper-evident and variable data areas are clear.

The sample data shows a print grade taken from the line template. In practice a grade is measured on a printed sample, and your setting holds it to the minimum you chose.

4

Returned with the reason already written

Approval stays locked while a difference is open. Annika returns version 4 to Rasmus Lind at the artwork house, copying Tomás Ferreira, the artwork coordinator. The reason is required and arrives pre-filled: the panel, the approved wording, the approved text version and the variation.

The reason also asks the artwork house to check that the 60- and 100-tablet cartons use the same side panel file, the line a reader focused on one file would forget. The dialog also shows what is at stake: the printer's slot on October 27 at Quendelhof Faltschachtel, with the new version needed by October 15.

5

Version 5 is read in full again, then signed

When version 5 arrives, all six agents run again from the first character: the planner confirms the storage line on side B, the extractor rebuilds 41 text blocks and 2,947 characters in reading order, and the comparer finds no differences. The run takes 4 minutes 12 seconds on the sample data, and the proof report is left ready for her signature.

The signature dialog shows the artwork, the version it was checked against, the result, and the meaning: "Artwork matches its approved text and may go to the printer." Annika enters her password and signs.

The solution's design target is about 3 hours per artwork proof, against about 2 days by hand. It is a target, to be measured on your own artworks.

A lost Braille number sign, a shrunken line, a stale date: the other errors in the wave, and the screens that track them

In shortThese screens show the other three kinds of error the wave catches, then the rollout by language and market, the source text, the checks across the wave, the signed record, the numbers over a year, and the rules you set yourself.

Select a screen to open it larger. The company, products, people, codes and figures are fictional. A rule's name on a screen is a pointer to the rule; the rule's real wording is in section 3.

Twelve months of proofs on the sample data

Six agents read, compare and measure every artwork; a labeling lead signs every proof

In shortEach agent has one job and limits it does not cross. None of them edits an artwork, dismisses a difference, passes a code below the set grade or signs. A difference closes in only two ways: a new version that passes, or an accepted deviation with a reason, approved by regulatory affairs. Nothing goes to a printer without a signed proof report.

Change wave planner to proof report writer: six jobs on one artwork

Labeling lead signs the proof; regulatory accepts a deviation; quality signs the press proof

Every artwork version, approved text version, proof and signature is kept, with agent steps and human decisions shown apart. The solution offers controls that support your compliance work under 21 CFR Part 11 and EU GMP Annex 11; your validation of its use stays yours. Botminds is SOC 2 Type II audited and manages information security to ISO 27001.

Annika, Tomás, Ines and Marek: the four people on the Ostravel wave

Print-ready artwork and approved text in; graded differences, checks and signed proof reports out

In shortThe solution reads the files your labeling chain already produces, and it leaves your artwork house, printers and systems of record in place. What it adds is one cited, signed report per approved artwork version, a rollout view of each change, and numbers you can hold each artwork house to.

Systems an artwork proofing rollout connects to

For each, check whether it has an API, a file feed, or only screens, and who owns it.

Your artwork house, your printers and your lifecycle system stay

Every difference cited and held open until closed, and every proof signed by a named person

What a hand-proofed artwork costs you in hours, reprints and missed print slots

In shortStart from your own costs, because the numbers that decide your case are yours. A few public figures frame them: how often labeling reaches a recall, how serious those recalls are, and how many packs a portfolio-wide change touches.

Timing ten proofs and pricing one reprint: finding your baseline in a week

Costs a proofreading calculator will not count

What faster proofs return when you count reader hours and late catches against the tool's yearly cost

In shortPut in your own numbers. The calculator asks how many artworks you proof a year, the hours each takes, what an hour costs, how many errors reach the press or later, and what the tool and your validation cost. No input has a public benchmark, so every default is marked illustrative or a design target.

Reader hours, returns and late catches: where artwork proofing saves

Artworks a year, proof hours and your cost of a late catch

If your proofs already take four hours with a good text-compare tool, your saving is far smaller than the default, and the case rests on Braille, codes, type size and the record. A company with a few dozen artworks a year probably should not buy this, and should say so before a demonstration.

How ready your labeling team is: artwork volume, approved text, severity rules, measurement and sign-off

In shortPick the answer closest to your team today. Eight questions score 0 to 2; the last is for reflection. The meter adds up your score as you go, and each question says why it matters and what a strong answer sounds like.

From 0 to 16: version your approved text, pilot on one product, or plan the portfolio

What to ask artwork proofing software vendors about outlined text, Braille, approved text and validation

In shortAsk each vendor, including us, to show the answers on your own carton. A demonstration on the vendor's file proves little. Look for three things: what happens to text it cannot read, what the system refuses to do, and what evidence it leaves behind.

Test artwork proofreading on your own cartons with planted differences, then decide on three numbers

In shortMeasure your current readers, plant known differences in real files, run them through both processes, and decide with three numbers. Plan about eight weeks on one product. None of these steps commits you to a vendor.

1. Time ten proofs, by step, and count versions per approved artwork over the last year.
2. Choose one real product: a multilingual carton, a leaflet, and one market with an awkward rule, with approved text as it exists today.
3. Plant known differences in retired copies: a digit, a look-alike character, an unchanged date, a line below the type minimum, a dropped Braille cell. Keep them away from anyone who tunes the tool.
4. Run both processes in parallel with the people who do the work, and with your people signing every proof.
5. Decide with three numbers: catch rate by error class, false alarms on clean files, and hours to a signed report.

Catch rate by error class, false alarms and hours to a signed report

A falling count of differences can mean better artwork or a weaker check, and the count alone cannot tell you which. Pair it with the severity mix and a seeded file every few months. A test with 20 planted errors of one class, all found, still allows a 14% miss rate (section 7), so size the test to the decision.

Artwork proofreading questions answered: two readers, Braille, Data Matrix, NDC and AI

In shortThe questions labeling teams and buyers search for most, each answered on its own.