Half of the Whitepaper's Bibliography Is About Timestamping Documents

Every argument about what Bitcoin was meant to be gets conducted without reference to the one part of the document that says where its ideas came from. The paper at bitcoin.org/bitcoin.pdf ends with eight numbered references. They take up a third of a page. They are almost never quoted, almost never followed, and they are the most straightforwardly informative thing in the file.

Here is what is actually in the list.

One entry is b-money, Wei Dai’s 1998 proposal. One is Adam Back’s hashcash write-up from 2002. One is Ralph Merkle’s 1980 paper on public key protocols, which is where the tree structure comes from. One is a 1957 probability textbook by William Feller, cited to support the arithmetic in the section on attacker success rates.

That is four. The other four are all about timestamping.

Reference two is a 1999 paper by Massias, Avila and Quisquater on designing a secure timestamping service with minimal trust requirements. References three, four and five all carry Haber and Stornetta: “How to time-stamp a digital document” from the Journal of Cryptology in 1991, “Improving the efficiency and reliability of digital time-stamping” with Dave Bayer added as first author in 1993, and “Secure names for bit-strings” in 1997.

So half the bibliography, and three of the eight entries carrying the same two names, is about a problem that has nothing to do with payments. The problem Haber and Stornetta were working on was how to prove that a document existed in a particular form at a particular moment, without anyone having to trust the party doing the proving.

What those papers actually built

The 1991 paper describes chaining. You hash a document, and you include in that hash the hash of the document that came before it. Do that continuously and you get a sequence where nothing can be inserted, removed or altered after the fact without breaking every link downstream. The value of any single entry comes from its position in the chain rather than from the reputation of whoever issued it.

The 1993 paper, with Bayer added as an author, improves the efficiency of that by batching. Rather than chaining every document to every other document, you collect the ones that arrive in a given interval, build a Merkle tree over them, and chain only the roots. That is why Merkle is in the reference list at all.

Their company published the current root value in a newspaper classified section on a weekly basis. The reasoning was that a printed newspaper is distributed to a very large number of independent holders at once and cannot be retroactively edited, which makes it a usable anchor even though no one involved has to trust the newspaper about anything.

Read those three papers in sequence and the architecture of the thing being proposed in 2008 is already most of the way assembled. Hash chaining, batching into trees, publishing the root somewhere that cannot be quietly rewritten.

What the paper calls it

Section three of the whitepaper is titled “Timestamp Server”. Not ledger, not currency, not payment network. The section opens by saying the solution begins with a timestamp server, describes it taking a hash of a block of items and publishing that hash, and states that each timestamp includes the previous one in its hash, forming a chain.

That is Haber and Stornetta, restated in a paragraph, and cited. What a timestamp is actually worth once the thing is running, and which rules keep a block’s claim about the time inside a range other participants will tolerate, is a separate question that the paper does not take up.

The material that people think of as the invention arrives in the section after it. Proof of work is introduced as the answer to a question the timestamping literature had left open: how to run this without a server. The timestamping papers assumed a service that clients submitted to. Section four replaces the service with a competition, and section six attaches a reward to winning the competition, and the reward is where the coin comes from.

The coin is the incentive that makes a serverless timestamp chain hold together. It is downstream of the timestamping problem, not the starting point.

Two of the eight are URLs, and both of them still answer

Six of the references are ordinary academic citations: a journal, a conference, a proceedings volume, a textbook. Two are bare web addresses, typed into the bibliography of a paper by somebody who evidently expected them to keep working.

Reference one is http://www.weidai.com/bmoney.txt. Reference six is http://www.hashcash.org/papers/hashcash.pdf. I fetched both while writing this piece, in September 2026, using nothing but a plain HTTP client.

Both return 200. Not a redirect to a rehosted copy, not an archive substitute: the same two files, at the same two addresses the paper printed, twenty-eight and twenty-four years after they were written. On a subject where a press release from 2014 is already gone, that is a remarkable result and it belongs on the record.

There is a catch, and it is the interesting half.

Neither address will load over an encrypted connection. weidai.com has nothing listening on port 443 at all, so a browser that has been told to prefer HTTPS gets a refused connection rather than a warning. hashcash.org does answer on 443 and then fails the handshake with a TLS alert. In both cases the document is reachable, and reachable only the way it was reachable in 2008, which is increasingly not the way a modern browser wants to reach anything.

The hashcash file carries its own provenance in its metadata, which is worth checking because it is the kind of detail nobody bothers to. Its PDF header gives a creation date of 1 August 2002 and names GNU Ghostscript 5.50 as the producer. The citation says 2002. The file says 2002. Those two facts were established independently and they agree, which is not something you can say about most of the dates in this subject.

The third URL in the document is the one on its own title page, www.bitcoin.org, and that one has been upgraded: a request for http://www.bitcoin.org/bitcoin.pdf today lands on https://bitcoin.org/bitcoin.pdf after a redirect. The paper’s own home got maintained. The two documents it was standing on did not, and are still there anyway, which says something about the relative durability of a maintained property and a file nobody has had a reason to touch.

Why this is worth the ten minutes

There is a durable argument about whether the design was meant to be money, infrastructure, a protest, or a store of value, and it is usually conducted by quoting the genesis block headline at each other. The bibliography settles less than people would like, but it settles something: the author was reading document integrity literature closely enough to cite three papers carrying the same two names, and structured the document so that the timestamping section comes before the currency.

The reference list is on the last page. It is not behind a paywall, it is not disputed, and it takes about a minute to read. It stays unread for the same reason most of the record in this subject stays unread, which is that a secondary account is quicker and there is no immediate penalty for relying on one.

Follow the citations. The paper tells you where it came from, in its own handwriting, and the answer is not the one that the last fifteen years of commentary would lead you to expect.

What was expanding, what was contracting

Run the dates in the list forward and one line is unmistakable. Merkle in 1980, Haber and Stornetta in 1991, the three of them with Bayer in 1993, Haber and Stornetta again in 1997, Massias and his coauthors in 1999. Two decades of continuous work on a single question, which was how to establish that a document existed in a particular form at a particular moment without trusting whoever says so. That line was expanding steadily, in journals and proceedings, and it was not about money at any point.

What was contracting, over exactly the same period and invisibly, was the addressability of the record it was all written into. The convention in that literature is a volume, a page range and a year, and a volume with a page range is findable in a century. The two entries in this list that broke the convention and gave a web address are the two that had no business surviving to be checked, and the sole reason the paragraphs above could check them is that two people, unpaid and unasked, left a text file and a PDF sitting where they put them. That is not a system. It is a coincidence repeated twice.

Who could tell at the time? Nobody had to. A reference list is not a prediction, it is a receipt, and this one was itemised. The author wrote down where the ideas came from, in the document, on the last page, for anyone who wanted to follow them, and what happened instead is that the field spent fifteen years arguing about the author’s motives while the page that answered the narrower and more answerable question sat unread underneath the argument.