The Price of Holding

On the evening of 16 October, Jameson Lopp posted four sentences about a spreadsheet. The archive of physical bitcoin attacks he has kept since 2014 had just crossed 700 entries. A researcher working under the name Pizza Bitcoiner had pointed an AI-assisted search at local-language newspapers, court records and police reports across 72 countries, and come back with 375 cases that had never entered the English-language record. Lopp's summary was characteristically dry: "Good news: the Bitcoin Wrench Attack archive just got a lot more accurate. Bad news: we just doubled the total number of known attacks to over 700."

Nothing in those 375 cases involved breaking a key. Nobody factored a modulus, nobody found a nonce, nobody exploited a wallet. Somebody was followed home, or visited at an address that appeared in a public filing, or held while a phone was unlocked. The oldest entry in the file is Hal Finney, SWATted in Santa Barbara on 29 December 2014, months before the block-size argument became a public fight.

That is the first thing worth saying plainly, because the industry has spent seventeen years arguing about the cryptographic perimeter while the documented losses accumulated somewhere else. But the second thing is the more interesting one, and it is not the headline. A record doubling in a day is not a crime wave. It is a measurement, and a measurement is the thing that converts a vague danger into a price.

What the file actually contains#

I downloaded the archive this morning and counted it myself rather than trusting the press release. The README holds 704 table rows. Two of them are column headers. That leaves 702 entries, exactly the announced figure. Of those, 688 carry a full date and 14 carry only a year or a phrase like "early January 2023"; assigning those to their stated years brings the total to 702 with no leftovers. The count is honest.

The shape inside it is where the argument starts. From 2014 to 2019 the yearly count runs one, five, four, twelve, twenty-seven, ten. From 2020 onwards it reads thirty-five, seventy-seven, eighty, sixty-six, ninety-nine, one hundred sixty-two. The current year stands at one hundred twenty-four through 17 October.

Look at 2019 to 2020 and you will see a step from ten to thirty-five and conclude that something changed. Something did, and it was not the attackers. Pizza Bitcoiner's database begins in January 2020. The archive's coverage boundary and that step are the same event. Anyone drawing a trend line through 2019 is drawing a chart of research method, not of violence.

Inside the comparable window the picture is less comfortable. From 2020 to 2025 the count rises from thirty-five to one hundred sixty-two, a factor of 4.6 on my division, with one dip in 2023. That is a rising series measured by a single method, and the 2025 number deserves a sentence of its own. The newsletter covering the announcement describes 2025 as a record year with "70-plus" documented incidents. The file now carries 162 rows dated 2025. Both can be true: the seventy-odd describes the archive before the additions, and the additions backfill years as well as weeks. What cannot be done is quoting the old figure and the new file in the same breath, which is exactly how a dataset gets misread by people who mean well.

Then 2026, which is the year nobody has framed yet. One hundred twenty-four entries cover 1 January to 17 October, 290 days. Annualised, that is 156, a fraction below 2025 rather than above it. The steepest part of this curve is behind us, or the reporting of it has saturated. Those two explanations are not distinguishable from inside the file, and I am not going to pretend otherwise.

Why a person in a doorway#

The mechanism of a wrench attack is the reason the numbers concentrate the way they do, and it has three parts. An attacker needs to know that a specific person owns bitcoin, needs to be able to tie that person to a name and a place, and needs the amount to be worth the trip. Modern cryptography defeats the first of those only if nobody talks, and the other two are not cryptography's department at all.

This is the joke in xkcd 538, which gave the genre its name: a cryptosystem defended with five-dollar wrench attacks against the person who holds the key. Sixteen years after that strip was drawn, it remains the most accurate threat model in the industry, and the archive is its bibliography.

The geography follows the same logic, and it is not what most readers would guess. France accounts for 85 of the 702 documented cases, 12.1 percent, the largest single country in the file. The United States accounts for 77, 11.0 percent. Taiwan accounts for 51, 7.3 percent. Brazil 35, England and Russia 32 each, India 29, Thailand 24.

Put those against population, using World Bank figures for 2024, and the concentration becomes strange. France holds 68.6 million people, 0.84 percent of the world's 8.14 billion. Its share of the archive is 14.4 times its share of humanity, on my division. The United States holds 4.2 percent of the world's population and 11.0 percent of the cases, a factor of 2.6. Taiwan's 51 cases amount to 7.3 percent of the archive; I have no World Bank population figure for Taiwan and will not invent one, so that ratio stays unstated.

Two explanations fit, and the archive cannot separate them. The first is documentation density: French-language press reports these crimes in detail, French courts publish, and the new pipeline reads French. The second is that bitcoin ownership in France is unusually visible, with a domestic industry that has spent years putting named founders in front of cameras. The newsletter's own figure for France in 2026 is "roughly 41" cases between January and September. Counting rows dated in that window gives 49. That difference of eight is itself a small lesson in how these numbers travel.

Three invoices for the same coin#

Here is the part that turns a security story into a market story. Every way of holding bitcoin already sends a bill. The bills differ in what currency they are denominated in, and only one of them appears on a statement.

The bank's invoice is the float. Mel Mattison, in an interview with TFTC published on 17 October, put Bank of America's stated cost of deposits at roughly 140 basis points while short-term rates sit far above that level, and described payroll processors holding billions in transit for three or four days as the same business in miniature. Treat the number as a guest's characterisation rather than an audited figure, because that is what it is. The mechanism does not depend on the precise spread: a deposit is a loan to a bank at a rate the bank sets, and the difference between that rate and what the bank earns on the money is the fee. It is invisible because it is charged in opportunity, and opportunity does not appear on a statement.

The fund's invoice is a management fee plus a right you do not have. Cory Klippsten's letter on 17 October states it in one line: "You can't withdraw your coins from an ETF or a Bitcoin stock." You own a claim on a custodian's stack, denominated in shares, settled inside the same plumbing as everything else you own. His letter doubles as a pitch for a share-for-coin conversion service, which is worth disclosing before quoting him, and it also contains the best framing I have seen for what that missing right is worth.

The self-custody invoice is denominated in the archive. No fee, no counterparty, no permission, and in exchange a documented, non-zero probability that someone will come to your house. Add the labour: multisig, geographic distribution of keys, a decoy wallet, the discipline of not discussing holdings, the years of not slipping.

Three invoices, three currencies: forgone yield, managerial fee, physical risk. The price of holding bitcoin is not one number, and the argument about which form is "better" is usually an argument between people paying in different currencies without saying so.

The sovereignty multiple#

Klippsten's thought experiment is the cleanest instrument anyone has built for pricing the third invoice. Would you trade two million dollars of a bitcoin ETF for one million dollars of coin you control? If yes, your sovereignty multiple is at least two. One and a half million? Then 1.5. He reports that more than half of the people he polled on X put their multiple above five. His own is around 1.5, as an American with access to the banking system and good lawyers. In a country with serious financial repression it would be ten. For friends in places where bitcoin is the only asset they trust, it might as well be infinity.

Read that as an exchange rate between a claim and possession, because that is what it is. It is also, unavoidably, a price on the third invoice: the multiple is what you are willing to pay in dollars to stop asking permission. What the multiple does not say is who you are protecting against.

That distinction decides everything. Against a state that can freeze an account, self-custody is the only instrument in the conversation, and a multiple of ten is cheap. Against a stranger who knows your address, the ETF is the only instrument that cannot be taken from you in your kitchen. The same person can hold both views honestly, because they describe two different attackers, and the mistake is to average them into a single preference about "custody" as though there were one threat.

The archive's contribution is that the second attacker is not hypothetical and no longer badly undercounted. Most bitcoin holders will never meet him. The one who does will discover that his keys were never the point, and that part of the sovereignty multiple he paid bought physical exposure rather than protection from it.

The case against this reading#

Three objections deserve more than a paragraph, because each of them could be right.

The first is the missing denominator. Seven hundred two cases over twelve years is a rate of roughly fifty-nine a year worldwide, and in the comparable 2020-to-2026 window about ninety-five a year on my division. Against a holder population in the tens of millions, that is a rounding error. I cannot state the rate per holder, because I have no sourced count of bitcoin holders, and inventing one to make the arithmetic look rigorous would be the exact failure this archive exists to correct. What the archive measures is not risk. It measures the number of people for whom the risk was realised and recorded, which is a different and smaller thing.

The second is inflation. An AI pipeline that searches 72 countries will also find duplicates, mistranslations, unverified local reports, and cases that are adjacent rather than on point. TFTC proposes the right test: if an independent audit finds duplication or out-of-scope inclusion above roughly twenty percent of the 375 additions, the doubling is an artefact. The 11 October curation, which added 38 entries, removed 3 and merged 33 duplicates, is evidence of discipline. It is not proof of accuracy, and the audit is the number to wait for.

Objection three: attacks may simply track the price. Rises in 2021 and 2025 sit on up-years, the dip in 2023 sits on a down-year, and 2026's plateau sits on a tape that has gone nowhere. Bitcoin traded at 81,608 dollars this morning on my own dashboard, 6.5 percent above the 76,600 dollar level at which The Bitcoin Layer's liquidity model last flipped to its red-dot signal on 11 September, a non-trade its own author describes as having missed a thirteen percent bounce at times. If the archive's annual shape is a lagged chart of price, then the 2020 step is method, the 2021 and 2025 spikes are bull markets, and the plateau is a flat market. That hypothesis is testable with annual average prices and I have not run it. Anyone who wants to argue with this essay should run it first.

What I could not check is worth listing, because the essay leans on it. There is no sourced count of bitcoin holders, so no rate per holder; the World Bank table carries no Taiwan figure, so no ratio for the third-largest country in the file; the 375 additions have not been independently audited; and I did not run the annual price series that the third objection would need to be settled. Each of those gaps is closeable, and none of them is closed here.

What to watch#

Three markers, each checkable rather than interpretive. The independent audit of the 375 additions, against that twenty percent threshold. The first insurer to publish a rate for physical bitcoin custody risk, because an underwriter's premium is the only place where this archive's contents become a price that markets are forced to read. And the next annual update: if the added cases land mostly in back years and the current year's count stays near 150, then documentation has caught up with reality and the series from here is a measurement of attacks. If the current year's count jumps again, the search is still finding history.

Consequence#

The interesting thing about a seventeen-year record in which nobody broke a key is what it says about where value actually sits. It sits in a person who can be found. Every case in the archive began with someone knowing who to visit, which means the load-bearing wall of self-custody is not the curve and not the seed phrase. It is privacy.

That is an uncomfortable conclusion for an industry that has spent a decade building transparency as a virtue. An address that can be tied to a name is a target with a postcode. A network that becomes more auditable as it becomes more valuable is charging its holders for the privilege, and the bill arrives in the currency nobody quotes on a screen. The archive just sent a second copy of that invoice, with 375 line items that had been sitting in Portuguese, French and Mandarin the whole time.

The price of holding was always there. What changed this week is that someone finally totalled it.

Sources#

• TFTC / Marty Bent — Lopp's Wrench Attack Archive Doubles to 702 After AI-Assisted Search
• Jameson Lopp — Physical Bitcoin Attacks archive, repository and dataset counted for this article
• Pizza Bitcoiner — bitcoin-wrench-attacks database
• xkcd 538 — Security, the five-dollar wrench
• Cory Klippsten / The Daily Bitcoiner — Would You Trade $2M of a Bitcoin ETF for $1M of Real Bitcoin?
• TFTC / Marty Bent — Mel Mattison: Banks Are Fleecing You
• The Bitcoin Layer / Nik Bhatia — TBL Liquidity Indicator Signal Update
• World Bank — Population, total (SP.POP.TOTL), 2024
• Bitcoin dashboard — own market feed, 18 October 2026, 09:31 CEST