The symbols an address can contain

An address of this kind is written in a restricted alphabet: the twenty six lowercase letters and six digits, 32 symbols in total. Nothing else can appear. No uppercase, no punctuation inside the string, and crucially not every digit.

The digits that are missing are the ones that look like letters. That is not an accident and it is the single most useful thing on this panel. If you are looking at something that claims to be a nexus market url and it contains a digit outside the allowed range, you can stop there. It is not a valid address, whatever else it might be.

The first tile draws all 32 allowed symbols as a grid. Each cell is shaded by how many of the three published addresses contain that symbol at least once. Fully shaded means all three carry it, mid shaded means two, faint means one, and an unshaded cell would mean a symbol that appears in none of them.

Across the three, 24 symbols turn up in every address. That sounds like a lot until you remember there are 56 characters in each one and only 32 symbols to choose from. With that much material, most symbols appearing somewhere is close to inevitable, and a set where they did not would be the surprising result.

The grid hides position entirely. Two completely different addresses can produce identical grids, so nothing about this picture identifies an address. It describes the ingredients and says nothing about the order.

The bar tile counts how many different symbols each address uses. The three counts are close but not identical, and the small differences carry no meaning at all. They are what you get when key material is written out in a small alphabet: some symbols happen to land more than once, some happen not to land at all, and which is which changes from one string to the next.

This is the first place on the board where it is worth resisting a story. A reader looking at three near identical bars will reach for an explanation of the differences. There is no explanation. The differences are the residue of randomness, and treating residue as signal is the most common way a dashboard misleads people.

The final tile counts symbols that appear in exactly one of the three addresses. Those are the closest thing the set has to distinguishing features, and they are still nearly useless for telling the addresses apart, because a symbol appearing once somewhere in 56 characters is not something a person will notice while reading.

It is included because it is genuinely computable and because it makes a point that recurs across this section: the properties that are easy to measure are almost never the properties that help you check something. Checking needs position. Every metric on this panel throws position away.

• A string containing a disallowed character is not an address. Fast, certain, and worth doing.
• A string using only allowed characters has told you nothing except that it could be an address.
• Symbol counts, coverage grids and distinct totals do not identify anything and should never be used as a fingerprint.
• The only check that identifies an address is a position by position comparison against a source you already trust.

Of the ten digits, only six are available here. The four that were left out are the ones that collide visually with letters in most typefaces, and removing them takes an entire family of transcription errors off the table before anybody has a chance to make one.

That is a small piece of design worth appreciating, because it is the opposite of how most identifiers are built. Ordinary reference numbers use the whole digit range and then rely on people being careful. This alphabet assumes people are not careful and removes the traps instead.

It does not remove all of them. Plenty of the remaining letters still trade places with each other under tired eyes, and the panel on confusable symbols counts how many of those survive in the published set. The lesson is only that the format has already solved the easiest part of the problem, and the part it left behind is the part you have to handle yourself.