Evaluation — April 2026

The NYT Investigation:
Strengths, Gaps,
and the Code Evidence

On April 8, 2026, the New York Times published John Carreyrou's 18-month investigation identifying Adam Back as Satoshi Nakamoto. Eric Ringger subsequently published a Bayesian model putting the probability above 99.99%. This page evaluates both — what the writing and behavioral evidence establishes, where it has limits, and how the code evidence fits in.
I The NYT Case at a Glance
Carreyrou's investigation is organized around four evidence categories: biographical alignment, a technical blueprint from the 1990s, linguistic analysis, and behavioral signals. It is the most rigorous Satoshi investigation published to date. It also never examined the source code.
Adam Back: British, 55. CEO of Blockstream. PhD in distributed systems, Exeter, 1996. Invented Hashcash proof-of-work in 1997. Active Cypherpunks contributor from 1995–2002. Has lived in Malta since 2009. Denies being Satoshi.
II The 1997–1999 Blueprint
This is the strongest part of Carreyrou's case — not similarities but a documented, multi-year technical roadmap that predates Bitcoin by a decade.
April 30, 1997: The Five-Attribute Framework
Very Strong

Back proposed an e-cash system "entirely disconnected" from banking with exactly Bitcoin's five core attributes: payer/payee privacy, distributed network resistance to shutdown, built-in scarcity, no required trust in any individual or bank, and a publicly verifiable protocol. All five became Bitcoin's foundation.

August 1997: Byzantine Generals and k-of-n Nodes
Very Strong

Back proposed a distributed bank where "k of n nodes have to collude before they can compromise" it — Bitcoin's exact security model. The phrasing maps almost directly: Back's nodes that "come and go" → Satoshi's nodes that "leave and rejoin the network at will."

December 1998: The Hashcash + b-money Synthesis
Uniquely Strong

After Wei Dai proposed b-money, Back suggested combining his Hashcash proof-of-work with b-money's coin minting. That is exactly how Bitcoin works. Nobody else on the Cypherpunks or Cryptography lists made this specific combination. Satoshi later described Bitcoin as "an implementation of Wei Dai's b-money proposal" — the combination Back proposed 10 years earlier.

A related wrinkle: the Satoshi emails Back produced for the Craig Wright trial show Satoshi only learning about b-money from Back in August 2008. But Back's Hashcash paper explicitly discussed b-money. Back conceded this was inconsistent. Carreyrou believes Back staged that email exchange to deflect suspicion — that he already knew b-money from his own paper, and pretended not to.

1998–1999: Inflation control, timestamps, energy defense
Strong

Back proposed increasing computational difficulty as chips improved — Bitcoin's exact inflation-control mechanism. He proposed public immutable timestamps using hash trees (Bitcoin uses this with Hashcash). He made the energy-defense argument using nearly identical language to what Satoshi would write a decade later.

Back (1999)
"As long as the wastage is lower than the costs of fiat money it's a win."
Satoshi (2009)
"If Bitcoin did grow to consume significant energy, I think it would still be less wasteful than the labour and resource intensive conventional banking activity it would replace."
Szabo proposed "bit gold" (never finished). Dai proposed b-money (didn't combine it with proof-of-work). Back assembled a near-complete Bitcoin specification across five posts over two years, including the specific combination — Hashcash + b-money — that Bitcoin actually uses. No other candidate comes close. This is the strongest piece of evidence in the investigation.
III Writing Analysis
Carreyrou used two separate methods: a professional computational linguist (Florian Cafiero) and an in-house AI-assisted word analysis with journalist Dylan Freedman. They reached different conclusions.
Cafiero Stylometry: Closest Match — But "Inconclusive"
Inconclusive

Cafiero compared the Bitcoin white paper against papers from 12 suspects. Back came out closest — but Finney was a very close second. After recomputing with a different distance metric, the results flipped. Cafiero called the results inconclusive — twice. He noted that if Satoshi knew how stylometry worked (and Back clearly did, having written about defeating it in 1998), it would be easy to game the analysis.

NYT Writing Funnel: 34,000 → 1
Strong (if methodology holds)

Starting with 34,000 mailing list users, filtered to 620 who discussed digital money, then applied 7 sequential writing filters:

  1. Two spaces between sentences → 562
  2. British spellings → 434
  3. Its/it's confusion → 114
  4. Sentences ending with "also" → 56
  5. "Bugfix" as one word, "halfway"/"downside" as one word → 20
  6. Incorrect hyphenation of "noun-based"/"file-sharing," correct "double spending" → 8
  7. Alternating e-mail/email, cheque/check, British/American "optimize" → 1: Back

Additional convergence markers: "proof-of-work" hyphenated as compound noun (8 people on the lists; Back is one); cross-referenced with "WebMoney" mention (4 people) — overlap yields Back only. "Partial pre-image" hyphenated — only Back and Finney used it; Finney wrote "preimage" in one word, Back hyphenated it, like Satoshi. "Burning the money" to mean destroying coins — only Back used this phrase pre-Bitcoin.

Hyphenation error analysis: 325 Satoshi hyphenation errors identified; Back shared 67 (next closest: 38).

Individual Phrase Parallels
Suggestive

Back (1996): "I'm better at coding, than constructing convincing arguments."
Satoshi (2008): "I'm better with code than with words though."

Back denied ever using "bloody." Carreyrou found a 1998 post where he used it. Back later denied Satoshi's 2015 block size email was a forgery, calling its observations "spot on" and "consistent with Satoshi views IMO" — then used the same vocabulary Satoshi used ("dangerous," "widespread consensus," "consensus rules," "trivial") in his own posts that same week.

The methodological problem: The filters were chosen because they appeared in both Back's and Satoshi's writing. You can't use the same observations to both select the filters and claim the filters identified the suspect — that's circular. It doesn't mean the result is wrong, but it does mean "34,000 to 1" overstates the independence of the finding. The honest denominator is 620, not 34,000, since only users who discussed digital money survived the first cut.
IV Behavioral Signals
Disappearance from Cryptography list (2008–2011)
Strong

Back was the most prolific commenter on e-cash proposals on Cypherpunks for a decade. During Satoshi's entire active period (late 2008 – April 2011), he made no public comments about Bitcoin on the Cryptography list. He claimed on the "Let's Talk Bitcoin" podcast that he "participated" in the 2008 discussion — no posts found. His first public Bitcoin comment was June 2011, six weeks after Satoshi's final post (April 26, 2011).

The Lerner Moment (April 17, 2013)
Strong

The day Sergio Demian Lerner published his analysis revealing Satoshi's ~1.1M coin fortune, Back joined Bitcointalk. Within hours he was proposing sophisticated improvements. Within two weeks he was demanding Wikipedia restore the standalone Satoshi page. Within 18 months he had founded Blockstream.

The 2015 Block Size War Vocabulary Match
Suggestive

Back wrote strident posts opposing larger blocks using specific vocabulary. Then "Satoshi" reappeared for the first time in 4 years using the same terms: "dangerous," "widespread consensus," "consensus rules," "trivial," "robust." Four days after Satoshi wrote "Very disappointing Gavin and Mike," Back wrote "Very disappointing Gavin and Mike."

The El Salvador Slip
Weak

Carreyrou: when he cited Satoshi's "I'm better with code than with words," Back interrupted with "I did a lot of talking though for somebody, I mean..." — phrasing Carreyrou interprets as Back speaking as Satoshi. Back denied it was a slip. The syntax is genuinely ambiguous; Carreyrou is reading a lot into a conversational stumble.

Metadata Refusal
Suggestive

Carreyrou requested the metadata for Back's Satoshi emails (produced for the Craig Wright trial). Back ignored the first email, ignored the follow-up eight days later, and never provided them. Even granting that the precautions Satoshi took would make the metadata nearly useless, the refusal is notable.

V The Ringger Model
Eric Ringger built a DAG Bayesian model with 10 evidence categories and Monte Carlo sensitivity analysis, arriving at a posterior above 99.99%. The model structure is sound. Three aspects of the inputs merit scrutiny.
The Prosecutor's Brief Problem
Critical Flaw

Every piece of evidence entering Ringger's model was sourced from Carreyrou's investigation — a one-sided inquiry. Bayesian models do not protect against bad inputs. A well-structured DAG with cherry-picked priors produces confidently wrong answers. The model has zero counter-evidence nodes, despite our code stylometry analysis being available at the time Ringger published.

The Stylometry Node
Misrepresented

Cafiero explicitly called his results "inconclusive" — twice, after two different computational approaches. Ringger's model assigns a high Bayes factor to the stylometry category, treating an inconclusive result as positive evidence. This is a direct misread of what the expert said.

Evidence Independence Assumption
Overstated

Many of Ringger's 10 evidence categories are not independent — they share underlying observations from the same mailing list corpus. The writing funnel, the blueprint posts, the behavioral silence, and the phrase parallels are all sourced from the same set of archived posts. Treating them as independent multiplies probability incorrectly, inflating confidence.

Three specific concerns reduce confidence in the 99.99% figure. First, every evidence input comes from a single investigation — there are no counter-evidence nodes. Second, the stylometry result entered as positive evidence was explicitly called "inconclusive" by the expert who produced it. Third, the code evidence is absent entirely. The extended Bayesian model incorporates the code and shows how it shifts the estimate.
VI The Code Problem
Neither Carreyrou nor Ringger examined Bitcoin's source code or compared it against Back's known output. That's the gap this site is built around.
Bitcoin v0.1: A Windows MFC Codebase
Counter-Evidence

Bitcoin's January 2009 release is unmistakably the work of a Windows GUI developer. Key identifiers:

  • C-prefix classes throughout: CBlock, CTransaction, CWalletTx, CKey, CAddress, CNode, CScript — the COM/MFC naming convention
  • Hungarian notation: nValue, nTime, hashPrev, vchPubKey, fRelay — systematically applied
  • Platform orientation: wxWidgets (cross-platform GUI), Windows registry access, Win32 API calls, precompiled headers (stdafx.h)
  • Code style: Allman bracing, // line comments exclusively (295 instances, zero block comments), C++ classes throughout (zero .c files)
Adam Back's Known Code: Unix C, Snake_case, Block Comments
1/12 Match

Back's Hashcash implementation and related academic code share almost nothing with Bitcoin v0.1:

Bitcoin v0.1
class CBlock { int nVersion; uint256 hashPrev; // line comment void SetNull() { nVersion = 0; } };
Back / Hashcash (C)
/* block comment */ int hashcash_check( const char *resource, int bits) { char token[MAX_TOK]; int version = 0; }

Back's code: C (not C++), snake_case (not CamelCase), block comments (not line comments), tabs (not spaces), no class hierarchies, no C prefix, no Hungarian notation, Unix-focused. Score: 1 out of 12 traits match.

Zero Cypherpunks Match Bitcoin's Code Style
Systematic Finding

Our survey of 30+ Cypherpunks mailing list participants with public code on GitHub found not one whose code resembles Bitcoin v0.1. The Cypherpunks community writes Unix C — open-source, command-line, platform-agnostic. Bitcoin was written by someone who thought in Windows MFC objects. These are two different developer cultures with minimal overlap.

This is the paradox at the center of the Satoshi mystery: the person who designed Bitcoin had Cypherpunk knowledge but Windows MFC coding habits. Back has the former but not the latter.

The obvious counterargument: Back wrote in 1998 about building tools to defeat stylometric analysis. Someone deliberate enough to game his writing fingerprint could deliberately code in an unfamiliar style too. That's possible. But adopting a foreign coding idiom across tens of thousands of lines — every class name, every variable, every brace placement — is a lot harder than adjusting prose style. More importantly, if you accept "any counter-evidence is intentional misdirection," you've made the theory unfalsifiable.

Assessment

Carreyrou built the most rigorous Satoshi investigation published to date. The 1997–1999 blueprint is the strongest single piece of evidence for any candidate — Back didn't just describe Bitcoin's ideas, he published a near-complete technical specification a decade before launch. The writing funnel is striking. The behavioral signals — radio silence during Satoshi's active years, joining Bitcointalk the day Lerner exposed Satoshi's holdings — are hard to explain away.

The code evidence complicates this picture. Back scores 1 out of 12 on code stylometry against Bitcoin v0.1. His Hashcash source, PhD work, and known academic code are Unix C — snake_case, block comments, no class structures. Bitcoin v0.1 is Windows MFC C++ — C-prefix classes, Hungarian notation, line comments throughout, Allman braces, precompiled headers. The divergence is substantial. Either Back is not Satoshi, or he wrote Bitcoin in a style deliberately foreign to his own practice. Both remain possible.

The extended Bayesian model (documented on the Bayesian Model page) incorporates the code evidence and still returns a high posterior, but with a meaningful uncertainty tail. The 99.99% figure and the code divergence are both real. They don't easily resolve into a single number.

The analysis most likely to shift the estimate: a direct code stylometry comparison of Back's post-2013 Blockstream C++ against Bitcoin v0.1 using quantitative tools. That comparison hasn't been done.

Evidence Summary: Back as Satoshi
Evidence Direction Strength Assessment
1997–1999 technical blueprint Pro Very strong Uniquely compelling — no other candidate matches this
Hashcash + b-money synthesis (1998) Pro Very strong Back was the only person to make this specific proposal
Writing funnel (34,000 → 1) Pro Moderate–strong Compelling, but filters chosen post-hoc from target's writing
Hyphenation analysis (67 vs 38) Pro Moderate Quantitative and verifiable; not decisive alone
Cryptography list silence (2008–2011) Pro Moderate–strong Hard to explain for the most vocal e-cash advocate
Lerner / Bitcointalk timing (2013) Pro Suggestive Same-day join has other explanations; notable regardless
2015 block size vocabulary match Pro Suggestive Four-day echo of "very disappointing" is striking
Metadata refusal Pro Suggestive Suspicious, but metadata may have been truly uninformative
Cafiero stylometry Neutral Inconclusive Expert's own description — twice. Cannot be used as positive evidence.
El Salvador "slip" Weakly pro Weak Ambiguous syntax; Carreyrou over-interprets
Code stylometry (1/12 match) Counter Moderate Bitcoin's MFC fingerprint doesn't appear in Back's code
Back's coding background (Unix/C, not Windows/C++) Counter Moderate PhD, Hashcash, career work: Unix-oriented throughout
Back's explicit denials Weakly counter Weak Expected regardless of truth; weak evidence either way
What would actually settle it: Compare Back's post-2013 C++ output (Blockstream, Liquid, sidechain work — all public on GitHub) against Bitcoin v0.1. If his C++ uses MFC-style patterns — C-prefix classes, Hungarian notation, Allman braces — the code counter-evidence largely collapses. If it continues to look like Unix C++ with no MFC influence, the code problem for the Back theory gets a lot harder to explain away.