Hal Finney Ran the Second Node and Wrote Down What He Thought

Almost everything we are told about the first months of this subject is somebody remembering. The memories arrive years afterwards, from people who by then know how the story turned out, and outcome contaminates recollection more thoroughly in this field than in almost any other, because the outcome was so large that nobody who lived through the beginning can pretend not to know it.

Hal Finney is the exception, and that is the whole reason he matters to a historian. He wrote his reasoning down while it was still reasoning, in public, on a mailing list with an archive, before there was anything to be right about.

This piece uses only his published writing.

November 2008: what he actually said

The whitepaper was announced on the Cryptography mailing list at metzdowd, and Finney replied on 7 November 2008. The reverent version of this story quotes the first paragraph:

Bitcoin seems to be a very promising idea. I like the idea of basing security on the assumption that the CPU power of honest participants outweighs that of the attacker. It is a very modern notion that exploits the power of the long tail. When Wikipedia started I never thought it would work, but it has proven to be a great success for some of the same reasons.

The reverent version tends to stop there. The message does not. Four paragraphs later:

Unfortunately I am having trouble fully understanding the system. The paper describes key concepts and some data structures, but does not clearly specify the various rules and verifications that the participants in the system would have to follow.

He then asks a series of specific questions about what a node does when it hears a transaction that does not make it into the next block, about what happens when a node is holding two candidate chains and a block arrives for one of them that conflicts with the other, and about how much history a recipient has to check. He closes:

This kind of data management, and the rules for handling all the packets that are flowing around is largely missing from the paper.

That is the first substantive technical critique of the design on record, and it came from its most enthusiastic early reader. Both halves are true at once and the second half is the one that usually gets dropped.

Six days later, on 13 November, he answered somebody else’s version of the same objection by working the mechanism out himself in the open:

Each node must be prepared to maintain potentially several “candidate” block chains, each of which may eventually turn out to become the longest one, the one which wins. Once a given block chain becomes sufficiently longer than a competitor, the shorter one can be deleted.

He is reconstructing mempool and chain reorganisation handling from a document that did not specify them, in November 2008, and getting the shape right. He also notes in the same message that the author had suggested releasing source code might be the clearest way to settle the question, and adds his own judgement: “perhaps writing real code is not a bad way to go.”

What he had already tried

He was not a bystander to the problem. In August 2004 he announced a system of his own to the Cypherpunks list:

I’d like to invite members of this list to try out my new hashcash-based server, rpow.net.

This system receives hashcash as a Proof of Work (POW) token, and in exchange creates RSA-signed tokens which I call Reusable Proof of Work (RPOW) tokens.

The interesting part is how he solved the trust problem, because it is the road not taken:

The RPOW server is running on a high-security processor card, the IBM 4758 Secure Cryptographic Coprocessor, validated to FIPS-140 level 4. This card has the capability to deliver a signed attestation of the software configuration on the board, which any (sufficiently motivated) user can verify against the published source code of the system. This lets everyone see that the system has no back doors.

RPOW keeps a server, and makes the server prove what it is running. The design he met four years later removes the server. A man who had built the first answer was the right person to see immediately what the second one was doing, and the record shows he did.

January 2009

He downloaded the first release. On the evening of 10 January 2009, on Twitter, he posted two words:

Running bitcoin

That message is the most reproduced artefact in this entire history and it is harder to read today than a 1995 press release. Requested from x.com without a browser, the page returns an application shell containing none of the text. The copy quoted above comes from an Internet Archive snapshot of the old twitter.com permalink taken on 12 December 2019, which also preserves the posting time as 7:33 PM on 10 January 2009. The primary source for the most famous sentence in crypto is a third party’s cache of a deleted web design.

The same evening he replied to the release announcement on the mailing list, congratulated the author, and raised the obvious practical difficulty: “there is still a difficulty in coming up with a reasonable argument in favor of a particular non-zero value for the coins”, noting “the practical problem that virtually no one will accept it at first”. The rest of that message is a speculative arithmetic exercise about what the currency would be worth in an imagined future, which is the single most quoted passage of his writing and the least useful. This site does not deal in that, and I am not relaying it.

On 24 January he answered a correspondent who argued that no government would tolerate the system, and his answer is the clearest statement anybody made that month about what had actually changed:

Bitcoin has a couple of things going for it: one is that it is distributed, with no single point of failure, no “mint”, no company with officers that can be subpoenaed and arrested and shut down.

Every earlier design in this cluster had officers.

The second node claim, and how much weight it holds

The title of this piece says he ran the second node. Here is exactly what that rests on, because the difference matters.

In March 2013 he posted a retrospective to the main web forum, and wrote:

When Satoshi announced the first release of the software, I grabbed it right away. I think I was the first person besides Satoshi to run bitcoin. I mined block 70-something, and I was the recipient of the first bitcoin transaction, when Satoshi sent ten coins to me as a test.

Note the hedges, which are his and not mine. “I think.” “Block 70-something.” He is describing events four years old, from memory, in exactly the way this piece opened by warning about, and he is more careful about it than most of the people who have quoted him since. The claim is his own, it is plausible, it has not been contradicted by anyone with a competing account, and it is a recollection rather than a record. Treat it as the strongest available evidence and not as documentation.

The same post says the software “made my computer run hot, and the fan noise bothered me”, which is the most human sentence in the early literature and a precise description of what running that program actually involved.

Two things this piece will not say

He is periodically named as the author of the whitepaper. That claim is not established, it has been advanced by others rather than by him, and he did not make it in any of the writing quoted here. Naming the claim is necessary because a reader will meet it; endorsing it is not available on the evidence.

And his own site is gone. rpow.net, the address in his 2004 announcement, still resolves, but what it serves today is an unrelated page in Indonesian about blockchain history, with no software on it. The 2004 announcement survives because it went to a mailing list somebody archived. His own publishing platform did not. That is the ordinary condition of sources in this period, and it is why the web forum carries a load it was never built for.

What was expanding, what was contracting

In November 2008, what was expanding was the written record: a design document, then a public argument about it, then a working reconstruction of the missing parts by a reader, all of it dated and archived.

What was contracting was the window in which anybody could write about this system without knowing the answer. It closed fast. Within a couple of years, every account of the beginning was being produced by people who already knew what had happened, and by 2013 even the participants were writing from memory.

Who could tell at the time? Finney could tell that the idea was strong and that the specification was incomplete, and he said both in the same message, six days apart, in public, with his name on it. That is what a contemporaneous source looks like, and it is the reason a historian trusts the questions in his November letters more than any confident answer written since.