Per address
| Address | Letters | Digits | Digit share |
|---|---|---|---|
| 5yymslrt···qum3apqd | 49 | 7 | 12.5% |
| 77x6lmfa···xgkmasad | 46 | 10 | 17.9% |
| 7iaj3wjp···muhcdjad | 51 | 5 | 8.9% |
| atmqbitg···fuj3usqd | 43 | 13 | 23.2% |
| dgkozv5m···345mopyd | 43 | 13 | 23.2% |
| evwigej4···npeuosad | 46 | 10 | 17.9% |
| fr6bvsrc···uams2tad | 50 | 6 | 10.7% |
| h3h66vql···ey6gslqd | 46 | 10 | 17.9% |
| jcyjjcu4···vogxwqyd | 46 | 10 | 17.9% |
| ncmebaso···hjlf44ad | 45 | 11 | 19.6% |
| otw35cxf···i7gt5lyd | 43 | 13 | 23.2% |
| torzon4v···qn76kiid | 45 | 11 | 19.6% |
The spread is the story
The set total sits within about one point of the flat expectation, which is unremarkable and exactly what you want to see. The individual strings do not. They run from five digits at the low end to thirteen at the high end, which as a share is roughly nine percent against twenty-three percent. Those two addresses were produced by the same process on the same day and they look like they came from different worlds.
This is the single most useful thing on the site for calibrating how much to read into a number. If a property can vary that widely between two strings that are equally valid, then that property cannot be used to sort valid from invalid. Anybody who tells you an address looks wrong because it has too many numbers in it is reading noise.
It also shows why aggregating helps. One string is 56 draws, which is far too few to settle near an expectation. Twelve strings are 672 draws, and the average lands close to where it should. Neither result is surprising. What is worth noticing is that the same underlying process produces a tidy answer at one scale and a messy one at another.
Why 18.75 percent
The alphabet holds 26 letters and 6 digits, which is 32 symbols. If a position is equally likely to hold any of them, the chance it holds a digit is six divided by thirty-two, which is 18.75 percent. Across 672 positions that predicts about 126 digits and the actual count is 119.
The reason the alphabet has exactly six digits is covered on the alphabet page. The short version is that four digits were removed to stop them being confused with letters, leaving 32 symbols, which is a convenient number because it encodes exactly five bits.
None of this changes what you should do with an address. It is context for reading the record pages, where digit counts appear as one field among a dozen, and where every one of them is a fact about a sample rather than a property of a key.
Questions
Is an address with lots of digits harder to fake?
No. Digits and letters cost exactly the same to grind. The alphabet does not treat them differently and neither does anybody generating addresses.
Why is the total not exactly 18.75 percent?
Because 672 characters is still a sample. Landing within about a point of the expectation is what a sample this size does.