Insights

39% of the First XRP Addresses Hide a Word. So Do 34% of Random Ones.

The account holding 80% of all XRP spells B1RD. David Schwartz’s founding account spells GoLD. A third reads b34r. There is a long-running belief that the founders encoded messages into the earliest addresses, and it is a reasonable thing to wonder. So rather than assert an opinion, we counted — every word in all 136 founding addresses, then the same count against addresses generated at random, to see whether the real ledger holds more of them than chance produces.

Every word we found, and where

Start with the data, because it is worth having either way. Fifty-three of the 136 founding addresses contain a four-letter English word — here is every one, largest holding first, with the matched letters highlighted where they sit. A means the match needed a leet substitution (1 read as i, 3 as e, 4 as a); the rest read straight off the address.

Address
Word
At
Held at 32,570
bird
7
79,997,608,218

Nineteen of the fifty-three are words a person would recognise as meaning something — bird, gold, bear, and, on an account holding ten million XRP, scam. The other thirty-four are mest, ichs, wied, seri, puru, balr, meum, jasz, aaru, yett and their kin — real dictionary entries that carry no meaning to anyone reading an address.

That ratio is the finding stated a second way. A generator hiding messages would produce messages; what is here is an alphabet doing what alphabets do. The genesis account itself, holding 201 XRP, spells jawy.

The claim is testable, which is rare

Most claims about intent are unfalsifiable. This one is not. An XRPL address is base58check over a 20-byte account ID, and if nobody is steering the process, the letters are effectively random. So the question has an exact form: do the 136 founding addresses contain more English words than random addresses do?

To answer it we need a fair comparison, which means generating fake addresses the same way the protocol generates real ones — twenty random bytes, a version prefix, a double-SHA256 checksum, and the XRPL base58 alphabet. The encoder checks out against a known value: account ID 1 encodes to rrrrrrrrrrrrrrrrrrrrBZbvji, which is exactly what rippled’s noAccount() returns. Then run 200 simulated ledgers of 136 addresses each and count.

What we found before we found the answer

Our first run said the riddles were real. It was wrong.
The first version of this test reported 3 hits observed against 0.4 expected, p = 0.015 — comfortably “above chance.” It was garbage, for a reason worth spelling out. The system dictionary was missing from the machine, so the script silently fell back to a hand-written word list. And that list had been written by us, earlier, from the three words we had already noticed— bird, gold, bear.

Searching for exactly the words you already found guarantees you will find them, and the probability attached to that finding is meaningless. This is the look-elsewhere effect, and it is not a subtle error. It is also, precisely, what riddle-hunting does: notice something, then go looking for more of it, then count the hits and not the misses.

The fix is to choose the dictionary before looking at the results, and to choose it large. We used 370,105 English words— 7,186 of them four letters long — and stripped the leading r, since every XRPL address has one and it is not a choice. Leet substitutions were normalised, so B1RD counts as bird and b34r counts as bear.

The result

Test
Observed
Expected
p
Observed vs expected
4-letter words
53 / 136
46.1
0.105
5-letter words
5 / 136
4.8
0.505
6-letter words
0 / 136
0.3
1.000
4-letter, no leet substitution
29 / 136
23.6
0.135

Blue is what the real ledger contains; grey is the average of 200 simulated ledgers. A result is conventionally interesting below p = 0.05. Nothing here is close.

53 of the 136 founding addresses contain an English word — and that is completely normal
Thirty-nine percent of the real addresses hold a hidden four-letter word. Random addresses hold one 34% of the time. The real ledger is seven addresses above a spread of plus-or-minus six, which is nothing at all. Words per address: 0.53 in the real set, 0.49 in the simulated one. Longest word found in either: four letters.

The test that should have found it, if it were there

A weak result could still hide a real effect on a few important addresses — if someone vanity-generated, they would have done it for the accounts that mattered, not for all 136. That predicts the large accounts should be wordier than the small ones. They are not:

Accounts ≥ 1bn XRP
29%
2 of 7 contain a word
Accounts < 1bn XRP
40%
51 of 129 contain a word

The effect runs backwards. The billion-XRP accounts are less likely to contain a word than the small ones. And of the six largest holdings in the ledger, four contain no English word at all — including Chris Larsen’s 8.189 billion, the contested 5.92 billion, and two of the billion-XRP founder grants.

So why does it feel like a pattern?

Because 39% is a lot. If you open a list of early XRPL addresses and start looking for words, you will find them in roughly two out of every five, and you will remember the ones that land somewhere meaningful. B1RD appearing in the account that held 80% of all XRP is genuinely striking — and there were 136 addresses and 7,186 four-letter words available to strike with. Something had to.

The misses are invisible, and that is the whole mechanism. Nobody circulates a screenshot of rJ6VE6L87yaV… containing sade, or r4mscDrVMQz2… containing herp, because those words mean nothing. Only the hits that land on a familiar concept get remembered, and a coincidence that lands well is indistinguishable, from the inside, from a message.

What would actually change our minds
This test can be beaten, and here is exactly how. Vanity generation leaves signatures: a shared prefix across related accounts, words appearing at a fixed position rather than scattered, words that are longer than five letters (the cost of generating those rises exponentially, so they are the tell), or a concentration of hits on accounts that share a funder. We found none of these. Show us a six-letter word in a founding address, or three related accounts with words at the same offset, and the arithmetic changes.

The same test, on a number instead of a word

The word-hunt is not the only version of this. A specific three-digit number, 589, has circulated in XRP communities for years as a figure with hidden significance. It is a good test case, because a number can be counted exactly the same way a word can.

There is a hook behind it, and it is worth dismantling carefully because we nearly repeated it ourselves. Price aggregators commonly report XRP’s early exchange price as $0.00589, and the digits match. That looks like confirmation until you check two things.

First, it is not XRP’s first recorded price. The earliest quote that survives is from 15 March 2013 at $0.00078 documented here, from a contemporaneous tracker whose own arithmetic reconciles. The $0.00589 figure is August 2013, five months later. Second, the aggregators disagree with each other: $0.0059, $0.00589 and $0.0067 all circulate for that period, so the third decimal that makes the coincidence work is not a settled number.

Which is to say the hook needs the wrong month and one particular rounding to land at all.

So we counted, the same way as before:

Where 589 could appear
Found
Expected
In any of the 136 addresses
0
In any account balance
0
In the 260 object indexes
3
~4
In the 260 PreviousTxnIDs
1
~4

And because a count is not evidence unless you can see it, here are the four:

A7CC9FE3F766F2DDE36D5530A0948C3BCB4DDB34D21B726091404E589D48AD07
C64C17E27388ED04D589D5537B205271B903C1518810602D50AD229FF74F11C5
F0F957EC17434D364BF0D48AC7B10065BFCFC4FEFC54265C2C5898C0450D85D9
4EFE780DF3B843610B1F045EC6C0D921D5EE1A2225ADD558946AD5149170BB3D
Three object indexes and one PreviousTxnID, at offsets 54, 17, 50 and 46 — all four on ordinary AccountRoots, none of them notable. Scattered, which is what chance looks like.

Four appearances against about four expected. A 64-character hexadecimal hash has 62 positions where a three-character run can start, and every character of “589” is a valid hex digit, so each hash has roughly a 1.5% chance of containing it — about four across 260 objects. That is what the ledger contains. The number does not appear in a single address or a single balance — and not in any of the four resting offers either, whose indexes begin 5B7F148A, 5F228268, 600A398F and D1CB738B. There is no 589 anywhere in the order book.

And ledger 589 cannot be looked at
The obvious place to check is ledger 589 itself. It falls inside the lost range, so no server on Earth holds it — the request simply fails. That is worth stating plainly because unfalsifiable is not the same as true. A claim that points at evidence nobody can retrieve has not been supported; it has been placed out of reach.

We are deliberately not going further than the arithmetic here. Some versions of this attach the number to guesses about who a particular pseudonymous account belongs to, and that is a claim about a real, named person — a different kind of claim, needing a different kind of evidence, and not one the ledger speaks to at all. What we can say is narrow and checkable: 589 appears in the founding ledger exactly as often as three random digits would, and the price coincidence is one number noticed after the fact, out of the very many numbers this history contains.

Reproduce it

Nothing here needs to be taken on trust. The state is public: read ledger 32,570 with ledger_data against s1.ripple.com and take the 136 AccountRoots. Generate the null set by encoding random 20-byte account IDs with the alphabet rpshnaf39wBUDNEGHJKLM4PQRST7VWXYZ2bcdeCg65jkm8oFqi1tuvAxyz under base58check. Use a dictionary you picked in advance, count both sets the same way, and compare.

If you get a different answer we would like to see it — but check first whether your word list was written before or after you looked at the addresses. Ours was written after, the first time, and it told us exactly what we wanted to hear.

Related: the oldest untouched position on the ledger · the lost ledgers · where the first 100 billion XRP went