Who Is Satoshi Nakamoto? A Bayesian Investigation

We got curious. Not the x-post-thread curious where you skim a few Reddit posts and call it research. Properly curious. The kind where you cancel your evening plans, open a fresh terminal, and decide you're going to quantify uncertainty like an adult.

The question, who is Satoshi Nakamoto?, has been answered a thousand times with varying degrees of certainty and approximately zero rigour. HBO made a documentary. Journalists have pointed fingers. Craig Wright has pointed at himself (we'll get to that disaster). But nobody, as far as we could tell, had actually sat down and done the maths properly. So we tried.

This isn't conspiracy theorising. This is Bayesian inference applied to one of the most fascinating pseudonymous identities in modern history. We're treating this like a founder would evaluate a market thesis: start with priors, gather evidence, compute posteriors, and see where the probability mass settles. If we're wrong, we'll update. That's the method.

The Setup: Why Bayesian?

Bitcoin's creator designed the system to be trustless. It seems fitting to analyse their identity the same way, without appeals to authority, without narrative bias, just evidence and probabilities.

We started with a uniform prior across ten candidates: nine named individuals who've been seriously proposed over the years, plus "Other" to capture the possibility of someone entirely uncatalogued. Each gets 10% to start. No favouritism. You earn your probability through evidence, not hype.

The candidates we considered (with apologies to anyone we've excluded, if you've got evidence for your pet theory, run your own model):

• Hal Finney – Cypherpunk legend, first Bitcoin recipient, creator of RPOW
• Nick Szabo – Bit Gold architect, smart contracts pioneer, obvious writing style match
• Adam Back – Hashcash inventor, cited in the whitepaper, UK-based
• Wei Dai – b-money creator, Crypto++ author, directly contacted by Satoshi
• Len Sassaman – Privacy advocate, remailer dev, died in 2011
• Paul Le Roux – Encryption genius turned cartel kingpin (yes, really)
• Craig Wright – Serial claimant, serial forger
• Peter Todd – Early Bitcoin dev, HBO's recent suspect
• Dorian Nakamoto – Wrong bloke, wrong evidence, included for completeness
• Other – The anonymous field

The Evidence: What We Actually Know

Satoshi left traces. Not many, but enough. We catalogued twelve types of evidence from primary sources: the Bitcoin whitepaper, 575 Bitcointalk forum posts, cryptography mailing list emails, the Bitcoin codebase itself, and various investigative analyses that have emerged over the years.

Here's what matters:

Age profile: Satoshi claimed to be born in 1975 on their P2P Foundation profile. That's 33 years old in 2008. Oddly specific if fake, though not impossible to fabricate.

Cryptography expertise: The whitepaper cites Hashcash and b-money. The code implements elliptic curve cryptography competently. This wasn't amateur hour.

C++ mastery: Bitcoin is written in C++, and written well for someone apparently coding solo. The architecture shows genuine software engineering chops, not just academic knowledge.

Economic and monetary theory: Satoshi understood inflation, deflation, Austrian economics, and game theory. Their forum posts reference gold standards and monetary policy with casual fluency.

Cypherpunk involvement: Active on mailing lists pre-2008, familiar with the community's privacy-first ethos.

Prior cited work: Direct references to Hashcash (Adam Back) and b-money (Wei Dai) in the whitepaper. Satoshi emailed Wei Dai in November 2008 asking about b-money and Bit Gold.

Location and timezone: Posts cluster around UTC and EST, not Pacific. British spellings appear ("colour", "optimise") mixed with American idioms. This suggests UK or US East Coast, possibly deliberate obfuscation.

Availability 2007-2011: Satoshi coded from mid-2007, published the whitepaper in October 2008, and disappeared in April 2011. They had to be alive, free, and engaged during this window.

Writing style: Formal, precise, patient, educational. No drama. Linguistic entropy analysis has been attempted by various researchers with mixed results.

Activity conflicts: Did the candidate have alibis during key Satoshi moments? Hal Finney was running a 10-mile race whilst Satoshi was sending emails and transactions in October 2009. That's rather inconvenient for the Hal-is-Satoshi theory.

Denials: Nearly everyone has denied being Satoshi. This is weak evidence either way—the real Satoshi would deny it, but so would everyone else.

Court-proven fraud: Craig Wright has been legally determined to be a forger and liar in UK courts. This is strong evidence.

The Model: Crunching the Numbers

For each candidate, we estimated the probability of each piece of evidence given that candidate is Satoshi. These aren't pulled from thin air, they're calibrated assessments based on everything we could verify: court documents, public CVs, archived mailing lists, code repositories, timezone analyses, writing samples, and credible investigative journalism.

Some examples to illustrate the thinking:

Hal Finney scores 0.9 for C++ (he wrote PGP implementations) but only 0.1 for "no activity conflicts" because of that race timing issue. His age (52 in 2008) scores 0.2 against the 1975 profile.

Craig Wright scores 0.1 for "no court-proven fraud", he's literally been ruled a fraud in court. He scores 0.1 for "has denied being Satoshi" because he's done the opposite, loudly and repeatedly.

Wei Dai scores 0.9 for cryptography (Crypto++ library), 0.9 for C++ (same reason), 0.9 for economics (b-money's design shows monetary sophistication), and critically 0.9 for "cited in prior work" because Satoshi directly referenced and contacted him.

We assumed independence between evidence types, an approximation, certainly (cryptography expertise and cypherpunk involvement correlate), but necessary to keep the model tractable. The posterior probability for each candidate is proportional to their prior multiplied by the product of all their evidence likelihoods, then normalised so everything sums to 100%.

The Results: Where Probability Settles

Here's what fell out when we ran the numbers:

Candidate	Probability
Wei Dai	33.53%
Adam Back	30.17%
Len Sassaman	15.77%
Nick Szabo	14.90%
Peter Todd	3.58%
Hal Finney	1.30%
Paul Le Roux	0.61%
Other	0.13%
Craig Wright	0.00%
Dorian Nakamoto	0.00%

Wei Dai edges it at 33.5%.

This wasn't the result we expected going in. The popular narrative favours Szabo or Finney. But the evidence doesn't lie, or rather, the evidence as we've weighted it points most strongly towards Dai.

Why? Several factors converge:

Direct citation and contact: Satoshi explicitly referenced b-money in the whitepaper and emailed Dai in November 2008. That's not circumstantial, that's documentary evidence of a connection.

Technical credentials: Crypto++ is a serious C++ cryptography library. This demonstrates both the coding ability and cryptographic knowledge Bitcoin required.

Cypherpunk bona fides: Active on the right mailing lists at the right time with the right ideology.

No disqualifying conflicts: Unlike Finney's alibi issue or Le Roux's cartel personality mismatch, Dai has no smoking gun that rules him out.

Subtle denial: Dai has never claimed to be Satoshi, but his denials have been understated rather than emphatic. Make of that what you will.

Adam Back comes in a close second at 30.17%.

Hashcash is cited in the whitepaper, he's got the cryptography background, he's UK-based (fitting the British English), and he was in email contact with Satoshi in August 2008. The model nearly splits the difference between him and Dai. The tie-breaker is Dai's superior C++ credentials and the fact that Satoshi's email to Back suggests they were separate entities ("You won't find anything" when discussing prior art).

Len Sassaman scores surprisingly well at 15.77%, largely because his 2011 suicide aligns eerily with Satoshi's final disappearance in April 2011. Some Bayesian AI models have pegged him at 99%, but we think that overfits a single timing coincidence whilst ignoring weaker economics and C++ evidence.

Nick Szabo, the perennial favourite, lands at 14.90%. Bit Gold is conceptually closest to Bitcoin, his writing style has been analysed as similar, and he's got the economic depth. But the C++ evidence is weaker, and he's been fairly emphatic in his denials.

Hal Finney drops to 1.30% primarily because of that October 2009 alibi—he was demonstrably running a race whilst Satoshi was active. That's a model-killer.

Craig Wright rounds to zero. The courts have spoken.

What This Actually Means

Let's be clear: 33.5% is not certainty. It's not even close to certainty. If you ran this model with different evidence weights or included candidates we missed, the numbers would shift. That's the point of Bayesian reasoning, it's a framework for quantifying uncertainty, not eliminating it.

Satoshi designed Bitcoin to be trustless and designed their own identity to be unprovable. We're not going to crack that with a spreadsheet. But we can say, with some rigour, where the probability mass sits given what we know.

If forced to bet; if this were a venture thesis and we had to allocate capital based on this model, we'd put our money on Wei Dai. Not because we're certain, but because the evidence, weighted and multiplied, tilts that direction more than any other.

Could we be wrong? Absolutely. The real Satoshi might be someone we've never heard of, someone who left no digital footprint beyond Bitcoin itself. The "Other" category captures that possibility, though our model rates it low because so much evidence clusters around the named candidates.

Why This Matters (Or Doesn't)

Bitcoin works whether Satoshi is Wei Dai, Adam Back, or a sentient algorithm running on a server farm in Iceland. The protocol is the point, not the personality.

But identities matter for different reasons. Satoshi's anonymity is a feature, not a bug, it prevents the system from having a single point of charismatic failure. The moment we know who Satoshi is, they become a target, a figurehead, a liability.

Still, we're curious creatures. We reverse-engineer puzzles because that's what humans do. This exercise wasn't about unmasking anyone or violating privacy. It was about taking a question that's been answered sloppily a hundred times and answering it properly, with maths, with evidence, with intellectual honesty.

If you disagree with our priors, fork the model. If you've got evidence we missed, update the likelihoods. That's how Bayesian reasoning works, it's collaborative, iterative, self-correcting.

We're not claiming to have solved the mystery. And if new evidence emerges tomorrow that flips the posteriors entirely, we'll update. That's the process.

---

Methodology note: Research constructed using LLMs. Evidence likelihoods calibrated from publicly available sources including court documents, archived mailing lists, academic papers, investigative journalism, and blockchain analysis. This is subjective quantification grounded in verifiable facts, not objective truth. Handle accordingly.