Making the Past Irrevocable

That is the whole situation. Everything else follows.

For as long as there have been records, a statement carried weight because producing it cost something: a person’s time, their name, someone who could be found and questioned. Remove that cost and you do not get more truth. You get statements that are no longer evidence of anything.

Information, then, is no longer scarce. Anyone can make it, in any quantity, in any voice. What is scarce is evidence that a particular event occurred.

The notary, the land registry, the patent office, the clearing house, the auditor, the court: ask what any of them is for, and you get the same answer in different clothes. What happened, and when. Who signed first. Who filed first. Who was told first. Who breached first. Strip away the buildings and the fees, and every institution of trust we have is a machine for answering that one question.

They answer it by being believed.

That is why they must be paid, and that is why they can be captured. The more the answer matters, the more it costs.

A record can answer the same question by being checkable instead.

---

A record is worth exactly its irreversibility. But irreversible with respect to what?

Not with respect to mathematics. With respect to someone who would have to do physical work to reach the material that would let them undo it. Against that adversary there are only two defences.

The first makes the work unaffordable. The material is still there, but reaching it costs more energy than the adversary has. This is what keeps a cryptographic hash from being reversed. It is proof-of-work. It is an economic wall, and it stands exactly as long as the adversary’s budget stays under the threshold.

The second makes the material disappear before the adversary arrives. Not expensive to reach: gone. You win a race, and afterwards there is nothing to buy.

There is no third way.

People often count three: work, destruction, and agreement. But agreement is not a mechanism. A blockchain’s past is irreversible only because rewriting it would cost more than it would earn; without that cost, consensus is a vote, and a vote can always be held again. Consensus is bookkeeping on top of the first wall, not a replacement for it.

The first wall has a price nobody prints on the box. Its security is denominated in the attacker’s budget; Bitcoin’s, in the price of bitcoin. A chain that refuses this coupling must decide in advance who may validate, and a list of validators is a polite name for power. Either way, a richer adversary erodes the wall, and security becomes a market variable.

That wall held while adversaries were merely rich. Three things now stand against it, and they are arriving together.

First, budget has stopped being a measure of capability. It was a good one while capability had to be bought. Machines that reason are already finding, unaided, flaws that no team had seen in some of the most heavily tested software in the world. In 2025, researchers extracted the attestation keys of secure enclaves in fully updated Intel and AMD server processors, with less than a thousand dollars of equipment. A wall priced in computation cannot hold against the thing that computes best. To trust it, you must require your adversary to stay poor, and then to stay slow.

Second, in 2026 a Federal Reserve paper (FEDS 2026-037) showed that where settlement must be paid for, its price rises precisely when everyone needs it at once, and that this rise turns caution into panic. The finding reaches beyond money: it is the property of every wall that is bought. Protection that must be bought is cheapest when no one needs it, and most expensive at the moment everyone does. An economic wall fails on the day it matters.

Third, the volume is no longer human. Machines are beginning to take actions that cannot be taken back, in quantities no institution was built to review, at speeds no reviewer can match. Their reasoning cannot be verified from outside, and this is not a limit of current tools: two systems that behave identically from the outside can never be told apart by any outside test. They can therefore be governed only by their effects, which is to say by their record. A record that defends itself only by out-computing the thing it records is no defence.

The second wall is denominated in nothing that can be bought. You cannot outbid the absence of a thing. You would have to beat physics, not a market.

It has one requirement, and that is where we should be attacked. Destroying something only makes it irreversible if it could not have been made twice. If the value can be recomputed from public inputs, a seed, an algorithm, a state anyone can rerun, then burning your copy achieves nothing. Nobody needed to steal it: it can be made again.

In software you can destroy your copy. You cannot make the thing unrepeatable.

That is what the hardware is for.

---

Every precision clock humanity has built is an oscillator and a counter. Pendulum, quartz, cesium: something swings, and we count the swings. But an oscillation is symmetric in time: film an ideal pendulum, play it backwards, and nothing looks wrong. Physics puts the arrow of time in exactly one place, the growth of entropy, and these clocks do not contain it. They measure duration and borrow direction from elsewhere.

Older clocks had the arrow. Sand does not climb back up the hourglass, a burnt candle does not grow back, decayed carbon-14 does not return. But they counted poorly, and nothing linked one instant to the next.

The hourglass had the arrow without the count. The pendulum has the count without the arrow.

As for the counter, the part that remembers what time it is, it is a register, and a register can be edited. Every digital clock ever made can be set back.

So they were never really clocks. They were watches: devices that tell the time without vouching for it.

Ours is built to join the two. Each tick consumes an unrepeatable physical event and chains it to the last. The direction is not displayed, it is enforced: no operation, at any level of privilege, runs it backwards, because the material that would let you re-strike a tick no longer exists. Before and after stop being stored values and become physical facts.

Not a better chronometer: a chronograph in the literal sense, not a time-measurer but a time-writer.

Writing time into a digital record is not new. Stuart Haber and Scott Stornetta did it in 1991, by chaining the hashes of documents one to the next, and they did it well; Bitcoin is built on their idea. But their chain writes time computationally, with no arrow of its own, and its irreversibility is that of the first wall, priced in an attacker’s budget. What this instrument adds is the thing that defines time in physics: one entropic event per tick.

Every digital clock until now told the time. This one writes it down.

---

A receipt seals four fields. The tip of the chain before it: a claim about history. The hash of the payload: a claim about content. The nanosecond: a claim about time. And the fingerprint of a physical event.

The first three can be false. The fourth can be neither true nor false.

It asserts nothing. It is not about the world; it is a piece of what happened in it. It has no truth-value. It cannot lie; it can only be irrelevant.

A signature is a claim. A fossil is not.

Signatures can be forged, because a signature says something, I was here, and everything that says something can be said falsely. A fossil says nothing. It is not testimony about an animal; it is what is left of one. You cannot forge a fossil by lying better. You would have to produce the animal.

Every security system built until now has tried to make claims harder to falsify. This one puts, inside the claim, something that is not a claim.

We built the whole thing around the one field that cannot lie.

---

That fourth field behaves unlike anything else in a computer, and the reason is simple.

Every digital object has a recipe. Given the program and the inputs, the bytes can be regenerated exactly: every file, every key, every random number drawn from software. All of it is made, and whatever is made can be made again. The event the instrument seals is not made: it happens. Its cause lies outside the digital world, in a measurement no one chooses, that never has an address, and that is gone before the receipt can be read.

Which is why the fossil was never a metaphor. A fossil is ordinary rock: it can be copied, cast, photographed. What cannot be copied is having been the animal. The bytes are the rock.

The origin is the animal.

One consequence belongs to no other system of trust. What was never held cannot be surrendered. A court can order the production of a key, and the order will be obeyed, because the key exists somewhere and someone holds it. The event cannot be produced on the order of any authority, because the material was consumed at the instant of its use. The operator is not relied upon to refuse. The operator is unable to comply.

---

This thread does not stop at engineering.

Augustine, sixteen centuries ago: What, then, is time? If no one asks me, I know; if I want to explain it to someone who asks, I do not know. Einstein, near the end of his life, told Carnap that the experience of the now troubled him: it means something essentially different from past and future, and this difference does not, and cannot, occur within physics.

The now is the one thing in the universe that cannot be given an address. Not because it is hidden: because it is everywhere. An address points at one thing among others, and the now is not among the things. It is the condition of there being any. Physics knows it only by its trace: the irreversible, what separates before from after.

Nor can it be named. A name is an address in language, a handle you take up again tomorrow to refer to the same thing. Reference is repetition. The now never repeats. The unrepeatable cannot be referred to; it can only be inherited.

This is where the instrument enters, and where the argument stays physics rather than mysticism. In the operational sense, a measurement is the irreversible registration of an event: the moment a possibility becomes a fact and cannot cease to be one. That is what experiencing the now is, stripped of poetry. The instrument does not name the present; nothing can. It does the only thing that can be done with a present: it lets it leave a mark, and it keeps the mark. The receipt is not a description of the moment, for descriptions are addressable, repeatable and forgeable. It is a consequence of the moment, and consequences are the only form in which the present survives itself.

In the digital world, everything has an address except the origin of that fingerprint. In the physical world, everything has an address except irreversibility itself.

The two worlds share exactly one orphan. We are building the door between them.

---

None of this makes anything true. A receipt proves that something was said, and when. Whether it is true remains the world’s business, as it always was.

---

The oldest thing in this argument is also the newest.

One of the oldest names writing has left us is Kushim, on a clay tablet from Uruk some five thousand years old. It reads, roughly: 29,086 measures of barley. 37 months. Kushim.

A receipt.

Writing was born to keep accounts. Literature, law and history came afterwards, almost as by-products. Civilisation did not begin with a poem. It began with someone keeping the books.

Scholars still argue whether Kushim was a man or an office. Five thousand years have dissolved the claim. Not the quantity. Not the duration. The record outlived what it was about.

That is the whole enterprise, written down before anyone could read.

We have one mission: to make the past irrevocable.