Length is the second cheapest check on this site after the alphabet, and it is the one people get wrong most often. Four ways to count, and the point at which each stops working.
Most of the shapes documented here are pinned by a number. Fifty six characters before the word onion. Sixty four hexadecimal characters. Twelve words, or twenty four. Get the number right and you have ruled out a whole class of damage; guess at it and you have ruled out nothing while feeling as though you had.
Nobody counts fifty six characters correctly by running an eye along them once. The methods below all work by turning one long count into a short count, which is the only thing that reliably survives contact with a real screen.
Group into fours or fives, then count the groups
Take the run in blocks of four or five characters and count blocks rather than characters. Fourteen blocks of four is fifty six. Nineteen blocks of five is ninety five. The arithmetic at the end is trivial and the counting has become a job of fourteen or nineteen items, which a person can hold.
Use whichever block size divides the target length cleanly, because a clean division gives you a second signal. If you are checking for fifty six and the last block comes out with two characters in it, you know at once that the string is not fifty six long without having finished the count.
Where the string is already grouped on screen, use that grouping rather than imposing your own. Fingerprints and seed phrases arrive grouped precisely so this step is done for you.
Use a known string as a ruler
Put a string whose length you already know on the line directly above the one you are checking, align the left edges, and compare where the right edges fall. This gives an answer without counting at all, and it is the fastest method on this page.
It only works in a fixed width font. In a proportional font an i occupies a fraction of the space an m does, so two strings of identical length end at visibly different points and two strings of different length can end at the same point. Before trusting a ruler comparison, confirm that the text is monospaced, which you can do by looking at whether a run of the letter i is as wide as the same number of the letter m.
The ruler you use should be a string you have a reason to trust, not a second copy of the string under test. Comparing something against itself only tells you that you pasted it twice.
Fold the run in half
Find the visual midpoint of the run, then compare the two halves for equal length by eye. Where the target length is even, the halves must match. Where they do not, the string is not the length you expect and you have learned that in one look, without ever producing a number.
Folding is also the quickest way to see doubling. A string that has been pasted twice looks, at the fold, like two runs that begin the same way, and the beginning of any string is the part people recognise fastest.
What the fold will not do is say which half is at fault. It flags the discrepancy and hands the work back to a proper count.
Count the separators instead of the groups
In a string that is already grouped, the separators are always one fewer than the groups. Five groups have four spaces between them; ten groups have nine. Separators are easier to count than groups because they are visually distinct from everything around them, and the eye lands on them without being directed.
This is the method to use on a grouped fingerprint or on a seed phrase, where the items themselves are long enough to blur into each other but the gaps are unmistakable. Count the gaps, add one, and you have the item count.
The trap is a trailing separator. A line that ends in a space, or a phrase pasted with a comma after the final word, breaks the relationship and gives you one group too many.
The lengths that matter here
| Characters | Counts as | What sits at that length |
|---|---|---|
| 32 | eight groups of four | A shorter hexadecimal digest, covered on a file hash |
| 40 | ten groups of four, or eight of five | The older fingerprint form, on a key fingerprint |
| 56 | fourteen groups of four | The body of the onion address, before the full stop |
| 64 | sixteen groups of four | A transaction id, and the longer digest on a file hash |
| 95 | nineteen groups of five | A primary Monero address |
| 106 | twenty one groups of five, then one over | An integrated address, on subaddresses and integrated addresses |
| 128 | thirty two groups of four | The longest hexadecimal digest on a file hash |
Four of the seven divide by four and two divide by five, which is the practical reason for keeping both block sizes in hand. The odd one out is 106, and the cleanest way to handle it is to count out 105 in twenty one fives and then notice the single character left over.
Where the method fails
Human counting is not accurate over long runs, and pretending otherwise makes the check worse than useless, because a wrong count produces a confident wrong conclusion.
- Recounting is not verification. If you count the same string twice you will tend to make the same mistake twice, in the same place, for the same reason. Two agreeing counts are one count.
- Length cannot separate two strings of the same length. Almost every substitution preserves length, so counting catches truncation, doubling and stray characters, and it does not catch a swap.
- A wrapped line is not a shorter string. When a run breaks across two lines the visual length of each fragment means nothing, and the fold method in particular will mislead you badly.
- Characters that occupy no visible column exist. Where one has been inserted, your count and a machine count will disagree, and yours will be the one that is wrong. This is the one failure the eye cannot be trained out of.
- Proportional fonts defeat the ruler, and some interfaces render a long string in a shortened form with an ellipsis in the middle. That is a display convention rather than a truncation.
A count that comes out wrong is worth acting on: something has happened to the string between its source and your screen. A count that comes out right has established that one number matched, which is the smallest thing a string can get right. The rest of that argument is on what a shape proves, and where it stops, and the faults a count will surface are indexed on wrong shapes, filed by what you see.