Chapter 2: The Price of Erasure

THE PRICE OF ERASURE

Thoth's Gift, Landauer's Limit, and the Physics of the Ledger

> This discovery of yours will create forgetfulness in the learners' souls. > — Plato, Phaedrus, in Jowett's translation

Introduction: A Being of Very Small Dimensions

In 1867, James Clerk Maxwell wrote a private letter to his friend Peter Tait describing a way to break the second law of thermodynamics. He was not trying to break it. He was trying to establish what kind of law it was, and he suspected — correctly, as it turned out — that it was a law about ignorance rather than a law about matter.

Here is the machine he imagined.

Take a box of gas at a uniform temperature. Divide it in two with a partition, and put a very small door in the partition. Now station at the door a being — Maxwell called it "a being whose faculties are so sharpened that he can follow every molecule in its course," and it was Lord Kelvin who later saddled it with the name demon — whose only job is to watch.

Temperature is an average. In any gas at any uniform temperature, some molecules are moving fast and some are moving slow. So the demon watches, and when a fast molecule approaches the door from the left, it opens the door and lets it through. When a slow molecule approaches from the right, it opens the door and lets it through. Otherwise the door stays shut.

The door is frictionless. Opening it costs nothing. The demon adds no energy to the gas; it does not push anything. It only sorts.

After a while, the right-hand chamber is hot and the left-hand chamber is cold. You now have a temperature difference where there was none, which means you have free energy where there was none, which means you can run an engine off it and do work, forever, for nothing.

This is not supposed to be possible. The second law of thermodynamics is the most reliable statement in all of physics — Eddington's line about it, that if your theory contradicts the second law there is nothing for it but to collapse in deepest humiliation, is quoted so often precisely because nobody has ever found an exception.

And yet nothing in Maxwell's setup is obviously wrong. The demon is doing no work. It is only looking, and deciding, and opening a door.

It took physics one hundred and fifteen years to find the flaw, and when it was found it was not where anyone expected. The demon does not fail at looking. The demon fails at forgetting.

That failure is the subject of this chapter, and the answer it produced is the single fact that this entire book rests on.

Section I: The Ledger

1.1 Thoth's Gift

In Plato's Phaedrus, Socrates tells a story he says he heard from Egypt.

The god Theuth — Thoth, ibis-headed, the scribe — comes before Thamus, king of Egypt, to present his inventions. He has invented number and calculation, geometry and astronomy, the games of draughts and dice. And last of all he presents writing, which he offers as a remedy for memory and wisdom.

The word he uses is pharmakon, and it means both remedy and poison. Greek did not distinguish. Neither, as it turns out, does thermodynamics.

Thamus is not impressed. He tells Theuth that the inventor is a poor judge of his invention's effects, and then delivers a criticism that has never stopped being relevant: writing will not strengthen memory but weaken it. Those who learn it will cease to exercise recollection, because they will trust to external marks instead of their own interior resources. They will hear many things and learn nothing. They will have the appearance of wisdom without the reality of it.

Every generation has re-run this argument with the technology updated. Writing. Printing. Calculators. Search engines. Whatever you are currently worried about your phone doing to your attention.

And every generation has treated Thamus as either an obvious reactionary or an obvious prophet, when the interesting thing about him is that he is neither, and he is describing a transaction rather than a decline.

Writing does weaken recall. That is not a slander on writing; it is measurable, and it is the point. Writing works by moving the burden of retention out of a small, expensive, biological store and into a large, cheap, external one — and the interior store, relieved of the burden, does not maintain a capacity it no longer needs. Something was gained and something was paid. Thamus saw the payment and mistook it for a swindle. Theuth saw the gain and did not mention the payment. Both of them were doing bookkeeping with one column.

1.2 The Weighing of the Heart

Thoth's other job, in the Egyptian material rather than the Greek retelling, is more literal.

In the Hall of Two Truths, the heart of the dead is placed on a scale against the feather of Ma'at — truth, order, the correct proportion of things. Anubis works the balance. And Thoth stands beside it with a palette and a reed, and he writes down the result.

Note the structure. The judgement is not the weighing. The judgement is the record of the weighing. And if the heart fails, Ammit — part crocodile, part lion, part hippopotamus — devours it, and the deceased is annihilated, which in Egyptian terms means something considerably worse than death: it means there is no longer anything to remember.

This is a ledger in the strict sense. There is an account. There is a scribe. There is a settlement. And the currency is not virtue in the abstract but a weight, a physical quantity, measured against a standard.

I am not going to claim the Egyptians anticipated statistical mechanics. `[ILLUSTRATIVE]` But I notice that when human beings imagine the ultimate accounting, they reliably imagine it as a written record of a physical measurement, with a settlement that cannot be avoided — and that this is, structurally, exactly what the second law of thermodynamics turned out to be.

1.3 Two Springs

One more, because it is the most compact statement of this book's thesis that anyone has ever produced, and it is about four thousand years older than the thesis.

In the Greek underworld there are two springs. One is Lethe, and the water of Lethe erases everything — the shades drink it and forget their lives, which is what makes them shades. The other is Mnemosyne, memory, mother of the Muses.

The Orphic gold tablets — thin sheets of gold leaf, buried with initiates in southern Italy and Crete, inscribed with instructions for the newly dead — tell the initiate exactly what to do at this junction. Do not drink from the first spring you come to. Go past it. Find the other one, guarded, by the white cypress, and say the correct words to the guardians, and drink from that one instead.

The initiate is being handed a compression policy. Not remember everything, which is not on offer, and not forget everything, which is the default. Choose your spring. Everything downstream of that choice — what kind of thing you will be, whether you persist at all — follows from it.

You cannot drink from both. That is not mysticism. As we are about to see, it is a constraint with a number attached.

Section II: The Demon's Wastebasket

2.1 The Wrong Answer, Held for Thirty Years

The first serious attempt on the demon came in 1929 from Leó Szilárd, who stripped Maxwell's apparatus down to the smallest version that still worked.

Szilárd's engine contains one molecule. That is the whole gas. You insert a partition into the middle of the box, which traps the molecule on one side or the other. Then you look, and see which side it is on. Then, knowing which side, you attach a piston to the appropriate wall and let the molecule push it as the gas expands isothermally back to the full volume, drawing heat from the environment and doing work.

Do the arithmetic and the work extracted is exactly kT ln 2, where k is Boltzmann's constant and T is the temperature. One bit's worth of knowledge — left or right — converted into that much work. Every cycle, forever, apparently for free.

Szilárd's resolution was that the measurement must cost at least that much. Looking is not free. Léon Brillouin developed the argument in the 1950s in terms of the photons you would need to bounce off the molecule to see it.

This was the accepted answer for thirty years, it is intuitively very satisfying, and it is wrong.

2.2 Landauer's Correction

In 1961, a physicist at IBM named Rolf Landauer was working on a question that looked purely practical: what is the minimum energy a computer must dissipate? Computers get hot. Everyone assumed there was some floor, and Landauer wanted to know where it was.

What he found was a distinction that nobody had drawn.

Some computational operations are logically reversible. If you know the output, you can uniquely reconstruct the input. A NOT gate is like this — flip a bit, and from the result you know exactly what it was before. No information is lost.

Other operations are logically irreversible. An AND gate takes two bits and returns one, and if the output is 0 you cannot tell which of three possible inputs produced it. Two or more distinct prior states have been mapped onto a single subsequent state. The distinction between them is gone.

Landauer's principle is this: *logically irreversible operations must dissipate energy into the environment, and the minimum is kT ln 2 per bit destroyed.* Reversible operations, in principle, need dissipate nothing at all. `[VERIFIED]`

The physical reasoning is not mysterious once you see it. The state of a physical memory occupies some volume in the space of possible configurations. If two distinguishable states become one, that volume has been halved — the system's entropy has decreased. The second law does not forbid a local entropy decrease. It forbids a total one. So the entropy has to go somewhere, and the only place available is the surrounding environment, and entropy delivered to an environment at temperature T is heat.

At room temperature, kT ln 2 is about 2.9 × 10⁻²¹ joules. Three sextillionths of a joule. It is the smallest bill in physics, and it is non-negotiable.

The cost is not in learning. The cost is in clearing.

2.3 Bennett Closes the Door

In 1982, Charles Bennett — also at IBM — took Landauer's principle and finally killed the demon, one hundred and fifteen years after Maxwell released it.

Bennett's argument runs like this. Grant Szilárd's critics their point: measurement can, in principle, be done reversibly and therefore for free. The demon looks at the molecule and copies the result into a blank register in its own memory. Nothing has been destroyed; a blank slot has been filled. That operation is reversible and costs nothing in principle.

The demon then uses that bit to extract kT ln 2 of work. Real gain, no cost yet. So far the demon is winning.

But now the demon's register is full, and the next cycle needs it empty.

The demon has a finite memory. It must, or it is not a physical object. And to run a second cycle it must return that register to a known blank state — which means taking a slot that could be either 0 or 1 and forcing it to 0 regardless. Two states to one. Logically irreversible. Cost: kT ln 2.

Exactly what it earned. To the last decimal.

The demon is not a fraud and not a magician. It is an engine that runs perfectly on a single tank of blank memory, extracts precisely as much work as its memory can hold, and then stops — unless it buys more blank memory, which is to say more unentropised substrate, which is to say fuel.

The demon's real resource was never information. It was emptiness. It needed somewhere to put what it learned, and clean space is a physical commodity with a price.

I would like you to hold that sentence for the rest of the book. Every archive in the universe — every star, every genome, every hippocampus, every civilisation — is in the demon's position. Not short of things to know. Short of somewhere to put them.

2.4 Somebody Measured It

A principle this consequential deserved a measurement, and for fifty-one years it did not have one, because kT ln 2 is an absurd quantity of heat to try to detect.

In 2012, a group led by Antoine Bérut, working with Sergio Ciliberto and colleagues at Lyon, did it. `[VERIFIED]`

Their bit was a single micron-scale silica bead suspended in water and held in a double-well optical trap — two adjacent laser-formed dips, one arbitrarily labelled 0 and the other 1, with the bead sitting in one of them. To erase the bit, they tilted and lowered the barrier so that wherever the bead had started, it ended up in the same well. Two states to one, physically enacted, with a bead you can watch under a microscope.

Then they tracked the bead's trajectory and computed the heat dissipated into the water.

The result: as the erasure was performed more and more slowly — approaching the quasi-static limit where the principle applies — the dissipated heat approached kT ln 2 from above, and did not go below it. Subsequent work using nanomagnetic bits and feedback traps has reproduced the result in other physical systems.

Landauer's limit is not a theorem about idealised computers. It is a floor in a beaker of water, measurable with a camera.

2.5 What the Principle Does Not Say

This is the chapter where a book like this can steal a great deal that does not belong to it. I would rather hand back the stolen goods in advance.

It does not say your laptop is running at the Landauer limit. It is running something like four or five orders of magnitude above it. The heat coming off a processor is overwhelmingly resistive and capacitive loss — charging and discharging gates through imperfect conductors — not the thermodynamic cost of erasure. Landauer's limit is a floor, and current technology is nowhere near the floor.

It therefore does not say that data-centre heat is the price of forgetting. You will see that claim made, and it is a lovely sentence, and it is not true in the way it sounds. Data centres are hot for ordinary engineering reasons. The Landauer contribution is negligible. `[VERIFIED]` I am refusing this because the honest version is more interesting: the limit tells us there is a floor at all, which is a statement about the universe rather than about Intel.

It does not say you can never compute for free. Reversible computing is a real research programme — Bennett, Edward Fredkin, Tommaso Toffoli — and a logically reversible machine can in principle run arbitrarily close to zero dissipation. But there is a catch, and the catch is the thesis of this book restated. A reversible computer never erases, which means it never throws anything away, which means it accumulates every intermediate result it ever generated. The garbage does not vanish; it piles up. Sooner or later you must either clear it, and pay, or acquire more blank tape, and pay for that instead. Reversibility does not exempt you from the ledger. It lets you defer the payment. `[SOURCED]`

And it does not, quite, end the philosophical argument. Landauer's principle is standard physics, taught and used and experimentally supported. But there is a live literature — John Norton and John Earman most prominently — arguing that the information-theoretic exorcism of Maxwell's demon is subtly circular, that it assumes thermodynamic facts in the course of deriving them. That dispute has not been resolved to everyone's satisfaction. The measurements are not in question. The claim that Landauer's principle is the final and complete answer to Maxwell is contested by serious people. `[BOUNDARY]`

I am building a book on this principle. You are entitled to know how firm the ground is under it: very firm as physics, less firm as philosophy, and I will not pretend the second thing is the first.

Section III: The Order That Costs

3.1 The Flame Is Not an Exception

There is an objection that occurs to everyone at this point, and it occurred to physics for a century before it was properly answered.

If entropy always increases, where do things come from? Galaxies are more ordered than the gas that made them. A tree is more ordered than soil and sunlight. You are, by a comfortable margin, the most intricately ordered object within several astronomical units. All of this happened after the Big Bang, in a universe supposedly running downhill.

The answer, worked out most fully by Ilya Prigogine and his collaborators from the 1960s onward, is that this framing has the relationship backwards. `[VERIFIED]` `[SOURCED]`

A system held far from equilibrium — with energy flowing through it, in one side and out the other — can spontaneously organise itself into structures that persist. Heat a shallow pan of oil from below and, past a threshold, the smooth fluid abruptly organises into a honeycomb of hexagonal convection cells, each one circulating, the whole pattern stable for as long as the heat flows. Nobody arranged the hexagons. They are Bénard cells, and they are what the fluid does when the gradient gets steep enough.

Prigogine called these dissipative structures, and the name contains the entire insight.

The hexagons are not resisting the second law. They are obeying it more efficiently than the smooth fluid was. An organised convection pattern transports heat from the bottom of the pan to the top faster than random conduction does. It produces entropy at a higher rate. Order emerged not despite the entropy gradient but because order was the faster route down it.

Schrödinger, in his 1944 lectures What Is Life?, had reached toward this with the awkward and much-criticised phrase about an organism feeding on negative entropy. He was right about the accounting and wrong about the metaphor. Living things do not consume order. They straddle a gradient — high-grade energy in, degraded heat out — and they maintain their internal structure by exporting disorder to their surroundings at a rate that more than compensates.

A candle flame is a dissipative structure. So is a hurricane. So is a whirlpool, a coral reef, a city, and every cell in your body.

3.2 Why This Belongs Here

Put Prigogine next to Landauer and you have the two halves of one statement, which is the reason this chapter exists.

Landauer tells you that destroying information costs energy. Prigogine tells you that maintaining structure requires an energy flow, and produces disorder somewhere else.

Together: there is no such thing as a free archive. Not a free acquisition, not a free clearing, not a free maintenance. A thing that persists is a thing paying, continuously, in a currency that ultimately comes out of the same gradient the whole universe is sliding down.

Which produces the law this book will keep returning to. I will state it once, plainly, and then use it for nine more chapters:

> Every persistent structure is a policy about what to discard, and the policy has a running cost. Nothing keeps everything, because keeping everything is not expensive — it is thermodynamically forbidden.

Notice how this changes the reading of Chapter One. When the fog cleared and the universe kept six numbers and lost a microstate, that was not the universe being tidy. The expansion did the work. The entropy went up. The compression was purchased, at the going rate, and the receipt is the microwave background falling on your face.

Section IV: The Budget

4.1 Twenty Per Cent

Your brain is roughly two per cent of your body mass and consumes roughly twenty per cent of your resting metabolic energy. `[VERIFIED]` It is, by a very wide margin, the most expensive tissue you own, and unlike muscle it cannot be powered down when idle — its baseline consumption barely differs between hard concentration and staring out of a window.

This is not a design flaw. It is what a dissipative structure of that complexity costs to run, and the cost is one of the strongest constraints in the whole of human evolution. Something had to be given up to afford it.

Which reframes the entire modern conversation about attention. We talk about focus as though it were a moral quality — as though a person who cannot concentrate has failed to want it enough. But attention is a selection operation performed by an organ with a fixed and severe energy budget, and every act of attending is simultaneously an act of not-attending to everything else. The filtering is not a side effect of attention. The filtering is attention. There is no version where you take it all in.

`[ILLUSTRATIVE]` — I am not claiming your subjective sense of concentration is Landauer erasure. I am claiming the constraint is of the same type: bounded capacity, mandatory selection, a real bill.

4.2 The Man Who Could Not Forget

In the 1920s a Moscow newspaper editor noticed that one of his junior staff, a man named Solomon Shereshevsky, never took notes at the morning briefing, and could nonetheless repeat the assignments back verbatim. He sent him to a psychologist. The psychologist was Alexander Luria, and he studied Shereshevsky for the next thirty years.

Luria could not find the limits of the man's memory. Not because they were large — because he could not locate them at all. Lists of a hundred nonsense syllables, recalled perfectly, forwards and backwards, and then recalled again without warning fifteen years later. Tables of random digits. Poetry in languages he did not speak.

And the man's life was, by Luria's account, a slow disaster.

Shereshevsky's recall was overwhelmingly sensory and associative — sounds had colours, words had textures and tastes, every stimulus dragged an enormous train of concrete imagery behind it. Which meant that abstraction was extremely difficult for him. He struggled to grasp the general point of a passage because every particular in it was blazing at full intensity. He had trouble recognising faces, because he had stored dozens of instances of each face at different moments and different lightings and did not naturally form the single averaged category the rest of us use. He eventually developed elaborate techniques for deliberately forgetting, most of which did not work. He ended his life doing stage memory performances, which he seems to have hated. `[SOURCED]`

Borges wrote the fictional version — Funes, who after an accident remembers everything and is consequently, in Borges's judgement, incapable of thought, because thinking requires the ability to ignore differences. He wrote it, as far as anyone knows, without having heard of Shereshevsky.

The lesson is not that memory is bad. It is that a category is a compression, an abstraction is a compression, a concept is a compression — and every one of them is purchased by discarding the particulars it summarises. A system that will not discard cannot generalise. It has an archive and no schema. It is a boat with every animal on it and no ark.

A note on the hyperthymesia literature. `[BOUNDARY]` A small number of people with highly superior autobiographical memory have been studied since 2006, and reports of intrusive and effortful recall are common in that group. The samples are tiny, the trait is heterogeneous, and I want to be careful not to turn living people with an unusual capacity into a parable. Shereshevsky is a single documented case study from a different century, and I am using him as one.

4.3 The Nightly Pass

If forgetting were merely decay — the slow rusting of unused traces — it would need no machinery. There is increasing evidence that it is not decay but an active process with dedicated mechanisms, which is a very different claim, because active processes are selected for.

Giulio Tononi and Chiara Cirelli's synaptic homeostasis hypothesis proposes that waking is a period of net synaptic strengthening, which is metabolically unsustainable and progressively degrades the signal-to-noise ratio of the network, and that a central function of sleep is a global downscaling — a broad weakening of synaptic weights that preserves relative differences while reducing absolute strength. What survives the pass is what was strong enough to survive it. `[SOURCED]` `[BOUNDARY]` — influential, well-supported in parts, and not the consensus in full.

Ronald Davis and Yi Zhong, and later Tomás Ryan and Paul Frankland, have pushed a related reframing: that forgetting should be understood as a form of learning rather than its failure — a regulated modification of stored engrams that keeps behaviour matched to a changing environment. `[SOURCED]`

A required caution. You may have encountered the claim that sleep flushes metabolic waste from the brain through a glymphatic system, based on work from 2013. It is a striking result and it has been widely repeated. It has also been directly challenged by more recent work reporting reduced rather than increased clearance during sleep, and the question is live. `[BOUNDARY]` I include it, flagged, rather than either using it or quietly dropping it, because a book about what survives a compression should not compress away its own inconvenient sources.

4.4 Thamus Was Half Right

Return to the king with the reed pen in front of him.

Thamus was right that writing weakens recall. He was right that there is a payment. What he could not see, standing where he stood, was what the payment bought — which was the ability to hold an argument longer than a human memory, to compare two accounts side by side, to catch a scribe's error four hundred years later, and eventually to run the entire enterprise that produced the measurement of kT ln 2 in a beaker in Lyon.

He also could not see that the alternative was not keeping everything. The alternative was a different and worse compression: oral transmission, which drifts, smooths, and reshapes its material toward whatever is memorable rather than whatever is true.

Neither option was lossless. Only one of them left a checkable trail.

That is the real distinction, and it is the one this book cares about. The question was never how much you keep. It is whether what you keep carries its own provenance — whether the summary is auditable back toward the thing it summarised, or whether it has cut itself loose and become a story.

The wooden people of the Popol Vuh had form and no memory of their makers. Thoth's writing had, for the first time, a form that could remember its makers. That is what the ledger is for.

The Four Slots

| Slot | The Szilárd engine / the demon | Human memory consolidation | |---|---|---| | Input | The molecule's actual position — one bit, freshly acquired into a blank register | A day of continuous multimodal experience, encoded at high volume into hippocampal traces | | Operator | Logically irreversible reset of the register: two possible states mapped to one known state | Active, regulated forgetting — synaptic downscaling and engram modification, substantially during sleep `[BOUNDARY]` | | Invariant | A demon capable of running another cycle: not the data, but the capacity to acquire more | Generalisable structure — schemas, categories, gist, procedural competence — retained at the cost of episodic particulars | | Cost | kT ln 2 per bit erased, released as heat into the environment, irreversibly `[VERIFIED]` | Roughly twenty per cent of resting metabolic energy to run the organ at all; loss of nearly all episodic detail |

The right-hand column is not a measurement and the left-hand one is. That asymmetry is the honest state of the field and the table shows it rather than hiding it.

The Protocol: Keeping the Books

`[ILLUSTRATIVE]` — application, not evidence.

Stop treating forgetting as a malfunction. It is an operation your nervous system performs deliberately, with dedicated machinery, at metabolic cost, because the alternative is Shereshevsky. If you have been running a low-grade grievance against your own memory for years, you have been blaming an organ for doing its job. The relevant question is not why do I forget so much — you forget so much because you must — but am I forgetting the right things.

Externalise deliberately, and pay attention to what the externalisation costs. Thamus was describing a real trade, not a decline. Writing a thing down genuinely does reduce the probability you will hold it internally. Sometimes that is exactly what you want — offload the reference material, free the expensive organ for the work that needs it. Sometimes it is not: the material you need to think with has to be inside, because you cannot form novel connections across an archive you have to look things up in. Decide which pile a given thing belongs in rather than defaulting.

Keep the provenance, not the volume. The failure mode of every note-taking system ever devised is that it becomes a second archive that no one can search, with the same information problem as the first one and none of the retrieval. The thing that makes an external record worth its cost is not its completeness. It is that a claim in it can be traced back to where it came from. A summary that carries its derivation can be decompressed. A summary that has lost its sources is a rumour with good formatting.

Budget attention as energy, not as virtue. The moralised framing — I should be able to focus on this — is bad physics and bad self-management. You are allocating a hard-limited resource across competing claims. That reframing does not make the allocation easier. It does make it a decision rather than a verdict on your character.

And accept the deferral for what it is. Reversible computing does not escape the ledger; it postpones the bill and accumulates garbage in the meantime. So does every version of I'll deal with this later — the unread inbox, the undecided decision, the unfiled year. Nothing has been avoided. The blank memory is filling up, and the demon cannot run another cycle until something is cleared.

Where This Leaves Us

• Maxwell's demon, proposed in 1867, appears to violate the second law by sorting molecules without doing work. `[VERIFIED]`
• Szilárd's 1929 single-molecule engine extracts kT ln 2 of work per bit of acquired information. His resolution — that measurement must cost that much — was accepted for decades and is not the correct one. `[VERIFIED]`
• Landauer's principle (1961) states that logically irreversible operations must dissipate at least kT ln 2 per erased bit; logically reversible operations need not dissipate. `[VERIFIED]`
• Bennett (1982) resolved the demon by locating the cost in the demon's need to reset its own finite memory, which is logically irreversible and costs exactly what the cycle earned. `[VERIFIED]`
• Bérut et al. (2012) measured the erasure cost in a colloidal double-well trap, finding dissipated heat approaching kT ln 2 from above in the quasi-static limit. Replicated in other physical systems. `[VERIFIED]`
• Current computing hardware operates several orders of magnitude above the Landauer limit; ordinary data-centre heat is not primarily the cost of erasure. `[VERIFIED]`
• Reversible computation defers rather than eliminates the thermodynamic cost, by accumulating unerased intermediate results. `[SOURCED]`
• Whether Landauer's principle constitutes a complete and non-circular exorcism of Maxwell's demon is disputed in the philosophy of physics literature. The experimental results are not disputed. `[BOUNDARY]`
• Systems far from equilibrium can spontaneously form self-maintaining dissipative structures, which produce entropy at a higher rate than the disordered state they replace. `[VERIFIED]` `[SOURCED]`
• The human brain is ~2% of body mass and consumes ~20% of resting metabolic energy, with little variation by cognitive load. `[VERIFIED]`
• Solomon Shereshevsky, studied by Luria over three decades, exhibited apparently unbounded recall together with marked difficulty in abstraction, categorisation, and face recognition. Single case study. `[SOURCED]`
• Forgetting appears to involve active, regulated mechanisms rather than passive decay. Specific proposals — synaptic homeostasis, engram modification, glymphatic clearance during sleep — are supported to varying degrees and the last is directly contested by recent work. `[SOURCED]` `[BOUNDARY]`
• The mapping from thermodynamic erasure cost to attentional and mnemonic economy is analogical. Shared constraint type — bounded capacity, mandatory selection, real cost — with no claimed shared mechanism. `[ILLUSTRATIVE]`

PART TWO

THE FOUNDRY

How structure gets made at all