The limits of text screening: three checks, not one

Monday, April 1, 2024, midday: adidas's own online store stopped letting anyone put a custom name and number on a German national team shirt. The German Football Association halted delivery of the number-44 orders it had already taken.

Timo, who runs a club's online shop, reads the news and at first cannot see why. A journalist on X had pointed out that the two digits, set in the shirt's typeface, resembled the SS rune. Bild picked it up, then the national press, as the Stuttgarter Zeitung laid out on April 2, 2024.

Timo opens his own shop's screening rules and types in 44. Nothing fires. Timo does not exist; his result does.

For any shop that sells personalization, that Monday is a pattern. A club shop is not dealing with one check but with three. The first asks what the text says, the second how the text looks once it is printed, and the third who is allowed to print it at all. Every case in which those three come apart has already happened.

The first case sits in nothing more than two digits.

Two digits walk past every rule

The hard part sits in the input field itself. "44" is two digits: no word list fires, no pattern looks unusual, no name-and-number rule triggers. Every text check a club or an agency could write would have let that order through. That is not carelessness: there was nothing in the string to find.

Four days later it was no longer about one combination. The federation had the digit 4 itself redrawn, as Sport1 reported on April 6, 2024. Its older three-dimensional 1990s form resembled the Siegrune, the emblem of the Deutsches Jungvolk in the Hitler Youth.

A text check reads a string. It never renders one.

None of those reports says what a stop like that costs. Timo can picture it.

Settled orders sit in the queue

A paid-for option that was selling on Friday is gone on Monday. Orders already settled sit in the queue and do not ship.

For the shop this is not one switch. The option disappears from the product page, every open order needs its own decision, and customer service explains the same thing for days. All of it lands in the weeks when a new shirt is ramping up.

The press made this case public. The three limits beside it make no headlines, yet they hit Timo's shop just as much.

Three more limits outside the text

Typography is the first, and the second is not in the field anyone types into.

Second list, not string

Some rules do not ask what is in the field. They ask whether it matches another source: is this number still in this squad? That is a lookup against a second list, and the pitch has run rules of this shape for years.

As an analogy, and strictly as one: UEFA's Equipment Regulations, in Article 11, require a printed player name to match the official player list. The name must be in a single color and in the Latin alphabet. That edition has been in force since June 1, 2024.

The rule binds the pitch, not the fan shop, but it shows the shape of the problem. And it moves: a kit-news blog reports that a new edition has applied since June 1, 2026.

A lookup that is correct today can be wrong next season. The third source is not even in a public document: it sits in a contract.

Sponsor contract, not word list

Whether a word collides with the shirt sponsor, or with a rival brand, is written in a contract, not in a word list. A club that changes sponsor mid-season has already changed what its own rule needs to say.

And even a name with nothing against it is sometimes one a shop may not print.

Rights question, not text question

In July 2025, UK sports media reported that a major English club's own store had stopped printing three former players' names. The photographed in-store sign gave the reason as "licensing restrictions". The SportBible report of July 24, 2025 rests on that sign, not on a club statement, so it counts here as reported.

None of the three names is offensive or blocked: whether a name may be printed at all is a rights question. A name that stops being valid mid-season is its own operational case.

None of this is legal advice. Where a name reaches into a competition regulation or a criminal-law boundary, it belongs with the club's own counsel, not with a screening tool. None of these three questions is answered by reading the string more closely.

Even where Timo's word list really is the right instrument, its reach stops somewhere.

Word lists stop at the language boundary

A word list is still the right first decision: it is cheap, it is easy to follow, and it catches exactly the input somebody built it for. It simply stops at a particular place.

A word list and a set of blocked combinations work only in the language they were built for. Folding disguised spellings, allowing ordinary surnames, grading severity: that curation is language-specific.

A configurator can show French labels on every screen and still have no screening coverage in that language. Interface language and screening language are two separate questions. The supplier checklist works through that split.

That leaves Timo's question: after all those deductions, what does his text check still deliver?

Text screening does one thing

A counter-example makes the limit sharp. Werder Bremen does not allow 18 or 88 as a custom flock. Its media director says so on the record, as the Deichstube reported on September 7, 2025. His reason: the numbers are "klare Codes aus dem rechtsextremen Bereich", established codes from the far-right scene.

Those two numbers are in the string, and a rule can catch them every time, with nobody watching. Whether a club writes that rule is its own call, and why the law does not make it for them is its own subject.

The 44 was in no string at all. It was a picture.

Everything a text check can do sits between those two cases. It does four things, every time:

• It compares strings with strings.
• It folds spellings together, so a list is not defeated by an inserted period, a swapped letter or a digit standing in for one.
• It checks the name, the number and the pair together, because a pair can be a problem when neither half is.
• It lets ordinary surnames through that would otherwise stick as partial matches.

kit.ink's own text screening works on exactly that principle. It evaluates text: names, numbers, free input, and the combination of name and number. Where a shirt carries several names and numbers, each one is checked. Typed signs such as runes, symbols and emojis are read as well.

Rendered shapes and images are not, and it knows neither sponsorship contracts nor competition regulations. So it does not see the case of April 1, 2024 either. That is not one product's gap. It is the edge of the category.

Inside that limit everything hangs on a second condition: which rule is live at all?

Rules go live one at a time

A check catches only what somebody activated a rule for beforehand. With no active rules it catches nothing, however good the mechanism underneath. Even a telling pair of name and number has to be written as a rule first.

Nothing in it learns on its own, and no model invents new bans by itself.

kit.ink's curated word lists cover eighteen languages today, at very uneven depth. German and English are the deepest, the others considerably thinner. Every imported rule lands as an inactive draft, and an import shows beforehand what it would change.

A new rule first runs quietly alongside real orders and shows what it would flag. Only then does a person at kit.ink switch it on, one rule at a time.

That step sits with kit.ink, not with the club, and a club cannot switch a rule off by itself either. It can propose one and look at what applies to it.

Everything beyond that machine half lands on somebody's desk.

A person decides the rest

A flagged name or number is held once the order arrives and put in front of a person, instead of being rejected outright. Shirts with nothing remarkable on them go through without a person.

While a case is held, no production system can read it. That stops the handover by kit.ink, and anything a shop exported before the hold sits outside that gate. A separate article covers the hold-versus-block distinction in full.

Of 66 official club shops read between August 4 and 6, 2026, 8 state concretely on their own pages when a personalization is refused. Each clause describes a decision made by hand, not a filter. Three of them name an "operator" who contacts the customer, as the full audit shows.

Where such a clause exists, a person is already the backstop for the cases that are not in the text. A tool that presents its word list as the complete answer asks less of that person than the shop's own published policy promises.

A tool that names its limit leaves that person the part only a human can do. That means looking at a printed shirt, checking a squad list, judging a rights question. The machine keeps the part it is good for.

Timo's shop has no such clause. It decides the same cases anyway, with nothing in writing behind it.

Three checks, not one

A text check answers one question well: does this string match a stored rule? It does not answer how the string looks in print, whether it matches a list held elsewhere, or who may print it.

Those are three separate checks. Which of them is written down in a given club appears on no shop page, not even on the eight that publish a refusal rule.

At midday on April 1, 2024 there was no error in any input field. The error was on the shirt. So Timo's test with the 44 was no failure of his rules. It only showed where they stop.

Anyone who keeps those three checks apart also knows where their own text screening ends and who takes over behind it.

---

To sort that split out for your own shop, start with a short conversation.

We accessed the Stuttgarter Zeitung and SportBible reports on August 6, 2026, the Deichstube report on August 5, 2026, and all other cited pages on September 22, 2026.