The annus mirabilis papers

Refiled. canon/proposals.md put this under ## idea and I have moved it to moment, on the project's own definitions rather than on taste. build.py says idea is "a framing later work is built out of rather than argued about. Cited when the thing being watched descends from it and the descent explains its shape." Nothing this project watches descends from special relativity or from the light quantum. No AI system's shape is explained by either. The two citation occasions the proposal itself names — the lone-genius narrative, and the argument about scale versus a better idea — are both about the circumstances of the year and not about any idea inside it, which is the tell. moment is "something that happened and changed what people expected next," and what happened is that one man with a full-time job in a government office sent five papers to one journal in six months. The unit here is a calendar, which is what the id says.

There is a second reason, and it is the harder one. Filing 1905 as an idea requires deciding which idea, and there is no answer. The five papers are in three unrelated subjects, and two of them point in opposite directions: the March paper argues that light is granular where Maxwell's theory says it is continuous, and the June paper takes Maxwell's continuous field so seriously that it rewrites space and time to preserve it. Einstein knew this and it bothered him for the rest of his life. An idea entry would have to smooth that over. A moment entry does not.

The proposal line, graded clause by clause. It reads: "four papers in one year from outside the academy, reorganising physics without new apparatus." Every clause is close enough to be dangerous.

"Four papers" — there were five, and the famous four are not the four he promised. In a letter to Conrad Habicht in May 1905 Einstein promised four: light quanta, the size of atoms, Brownian motion, and "an electrodynamics of moving bodies." The set now called the annus mirabilis papers drops the second (the doctoral dissertation) and adds a fifth he had not thought of yet, the three-page mass-energy note of September. Two different fours. Section 1 sets out all five.

"In one year" — six months and nine days, from 18 March to 27 September 1905, by date of receipt at Annalen der Physik. The dissertation did not appear in that journal until 1906.

"From outside the academy" — half right, and the wrong half is the half people lean on. He held a physics teaching diploma from the ETH Zurich, had published five papers in the same journal between 1901 and 1904, and submitted an accepted doctoral thesis in the same year. He was not an amateur and he was not a stranger to the editor. What he genuinely lacked was a university post, a library, a supervisor, a laboratory, a grant, and a single colleague working on his problems. Both halves are citable and they are citable for opposite purposes.

"Reorganising physics" — right, and not in 1905, and not for years. The best experiment then available reported in 1906 that special relativity was refuted. His application to lecture at the University of Bern was rejected in October 1907. He was still examining patents in July 1909. The Nobel came in 1922, and its citation deliberately excluded relativity. Section 1 carries the reception in detail because the compressed version of it is the single most misleading thing about this entry.

"Without new apparatus" — right, and it is the sharpest clause in the proposal and the reason the entry is worth having. It also needs its other half stated in the same breath: confirming those papers took a great deal of apparatus belonging to other people over eighteen years, and one of the five was wrong as published until somebody else's laboratory caught the error. Sections 2 and 3 are built around that pairing.

The kinds I declined.

limit was the real alternative, and maxwell-1865 — already written — explicitly routes here for it: "the claim that nothing whatever can outrun that speed is relativity's, and einstein-1905 is proposed and unwritten. The canon has no entry for it yet." That deferral is honoured, in section 4, which is where a limit entry's required content lives. I did not change the kind to collect it, because the light-speed bound is one result out of five and refiling the whole year under it would misrepresent four papers to accommodate one. If a later job wants a dedicated limit entry for the invariance and boundedness of c, it should write one and point it here; this file gives it the misuse catalogue to start from.

prediction fails on the kind's own rule. This canon's prediction entries exist to build a base rate for how wrong AI forecasting runs, and they are required to carry the predictor's grade and an independent one and say the two differ. Einstein made no dated forecasts of the sort — the closest thing is the September paper's suggestion that radium salts might test the mass-energy relation, which is a proposed experiment rather than a claim about the future. His one recorded self-grade is late and negative and appears in section 3.

interpretation would be a category error. Nothing here is a reading of events.

descends_from: [maxwell-1865], and the descent is documented in the first sentences of two of the five papers. "Zur Elektrodynamik bewegter Körper" opens on an asymmetry "in Maxwell's electrodynamics — as usually understood at the present time." The light-quantum paper opens on "a profound formal difference" between the theory of gases and "Maxwell's theory of electromagnetic processes in so-called empty space." Maxwell is also one of only five scientists named anywhere in the relativity paper — Newton, Maxwell, Hertz, Doppler and Lorentz. The other genuine ancestors are Lorentz's 1904 transformations, Poincaré's work of 1900–1905, Planck's 1900 quantum, and Boltzmann's statistical mechanics. None of them is in canon/ or canon/proposals.md, so there is no id to name and I have invented none.

One cross-reference worth setting before it is set wrongly. time-machine-1895 already carries the correction that Wells did not anticipate Einstein, and it is right; the fourth dimension in that novel is a narrative device and has no relativistic content. This entry does not overturn that and should not be cited to.

What it is

Who he was in January 1905. Albert Einstein, twenty-five, was a salaried employee of the Swiss Federal Office for Intellectual Property in Bern. He had been appointed in June 1902 and started work on 23 June 1902 as a level III technical expert at 3,500 francs a year; he was promoted to level II on 10 March 1906 at 4,500 francs, and resigned on 16 July 1909 to take an associate professorship at Zurich. The patents he examined were gravel sorters, meteorological instruments, electric typewriters and motors. He was married to Mileva Marić, who had studied physics alongside him at the ETH, and they had a young son. He read in the evenings with two friends, Maurice Solovine and Conrad Habicht, in a self-mocking circle they called the Olympia Academy. His closest technical interlocutor was another patent office engineer, Michele Besso, of whom he later said he "could not have found a better sounding board for my ideas in all of Europe."

He was not unpublished. Between 1901 and 1904 Annalen der Physik carried five Einstein papers — on capillarity (AdP 4, 513), on potential differences between metals and salt solutions (8, 798), on the kinetic theory of thermal equilibrium (9, 417), on the foundations of thermodynamics (11, 170), and on the general molecular theory of heat (14, 354). None made a mark. He was a minor working physicist with a day job.

The letter. In May 1905 he wrote to Habicht, who had failed to send his own dissertation, in the register the Olympia Academy used with each other:

> "So, what are you up to, you frozen whale, you smoked, dried, canned piece of > soul…? Why have you still not sent me your dissertation? Don't you know that I > am one of the 1.5 fellows who would read it with interest and pleasure, you > wretched man? I promise you four papers in return. The first deals with > radiation and the energy properties of light and is very revolutionary, as you > will see if you send me your work first. The second paper is a determination of > the true size of atoms… The third proves that bodies on the order of magnitude > 1/1000 mm, suspended in liquids, must already produce an observable, random > motion that is produced by thermal motion… The fourth paper is only a rough > draft at this point, and is an electrodynamics of moving bodies which employs a > modification of the theory of space and time."

Two things in that paragraph matter more than the physics. The one he called "very revolutionary" is the light-quantum paper, not relativity — and it is the one the profession refused for eighteen years. And relativity, the thing his name now means, gets one flat clause at the end and is described as a rough draft. He was right about which paper was the outlier and wrong, along with everybody else, about which one would land.

The five papers. All dates are dates of receipt at Annalen der Physik unless stated. The journal ran two volume numberings at once — the fourth-series numbers (17, 18, 19) and the whole-series numbers (322, 323, 324) — which is why citations of these papers disagree in a way that looks like error and is not.

1. "Über einen die Erzeugung und Verwandlung des Lichtes betreffenden heuristischen Gesichtspunkt" — On a Heuristic Viewpoint Concerning the Production and Transformation of Light. Received 18 March 1905, published 9 June, AdP 17, 132–148. It opens: "Between the theoretical conceptions which physicists have formed about gases and other ponderable bodies, and Maxwell's theory of electromagnetic processes in so-called empty space, there exists a profound formal difference." Matter is counted; light is a continuum. Einstein proposes taking the granularity seriously: that the energy of light "is not continuously distributed over an ever-increasing volume, but consists of a finite number of energy quanta localized at points in space, which move without dividing and which can only be absorbed or emitted as wholes." From this he derives, among other things, that the maximum kinetic energy of an electron ejected from a metal by light is the quantum energy minus a work function — independent of the light's intensity and linear in its frequency. The word "heuristic" in the title is doing real work: he is not claiming to know what light is, he is claiming that assuming this produces results.

2. "Eine neue Bestimmung der Moleküldimensionen" — A New Determination of Molecular Dimensions. The doctoral dissertation, submitted to the University of Zurich on 20 July 1905 under Alfred Kleiner, who returned it once as too short. Published in AdP 19, 289 in 1906. It relates the viscosity of a liquid with and without suspended particles, applies the relation to sugar dissolved in water, and extracts a value for Avogadro's number and a size for the sugar molecule. It is the least famous of the five and, because of its use in cement, dairy, aerosol and suspension engineering, it is the most cited paper Einstein ever wrote.

3. "Über die von der molekularkinetischen Theorie der Wärme geforderte Bewegung von in ruhenden Flüssigkeiten suspendierten Teilchen" — On the Motion of Small Particles Suspended in a Stationary Liquid. Received 11 May 1905, published 18 July, AdP 17, 549–560. If matter is made of molecules in thermal motion, then a particle small enough will be visibly kicked around by them, and the statistics of that kicking are calculable: the mean squared displacement grows linearly with time, with a coefficient fixed by temperature, viscosity, particle size and Avogadro's number. Einstein noted — cautiously, and in the paper — that what he was describing sounded like the jitter biologists had been calling Brownian motion since 1827. The importance is not the jitter. It is that atoms were still a hypothesis in 1905, seriously disputed by serious chemists, and this turned them into something a microscope and a stopwatch could count.

4. "Zur Elektrodynamik bewegter Körper" — On the Electrodynamics of Moving Bodies. Received 30 June 1905, published 26 September, AdP 17, 891–921. It opens on the magnet and the conductor: move the magnet or move the coil and you get the same current, but the textbook explanation of 1905 gives two completely different accounts of why, and "the observable phenomenon here depends only on the relative motion of the conductor and the magnet, whereas the customary view draws a sharp distinction between the two cases." Add "the unsuccessful attempts to discover any motion of the earth relatively to the 'light medium'" and Einstein takes both as evidence for two postulates: that the laws of physics are the same in all frames in uniform translation, and that light moves at a definite speed c in the stationary frame "whether the ray be emitted by a stationary or by a moving body." From those two, with no new experiment and no new mathematics, follow the relativity of simultaneity, the Lorentz transformations, time dilation and length contraction — and the conclusion that "the introduction of a 'luminiferous ether' will prove to be superfluous."

The paper cites nothing. No references, no bibliography, no engagement with the literature it is answering. Michelson–Morley is not named; Lorentz's 1904 memoir is not named; Poincaré, who was working the same ground the same month, is not mentioned at all. Five scientists appear anywhere in it, and only as adjectives: Newton, Maxwell, Hertz, Doppler, Lorentz. It closes on a sentence of acknowledgment: "In conclusion I wish to say that in working at the problem here dealt with I have had the loyal assistance of my friend and colleague M. Besso, and that I am indebted to him for several valuable suggestions."

5. "Ist die Trägheit eines Körpers von seinem Energieinhalt abhängig?" — Does the Inertia of a Body Depend Upon Its Energy Content? Received 27 September 1905, published 21 November, AdP 18, 639–641. Three pages, no citations except to his own June paper. "If a body gives off the energy L in the form of radiation, its mass diminishes by L/c²." Generalised: "The mass of a body is a measure of its energy-content." And then, uniquely among the five, a proposed experiment: "It is not impossible that with bodies whose energy-content is variable to a high degree (e.g. with radium salts) the theory may be successfully put to the test."

What happened next, which is the part that gets compressed away.

Relativity was contested immediately, and for two years the best measurement said it was wrong. Walter Kaufmann had been deflecting fast electrons since 1901 to distinguish between competing models of the electron. In 1906 he published results finding Abraham's and Bucherer's models inside his experimental error and the Lorentz–Einstein prediction outside it, and concluded that Lorentz's theory was refuted and Einstein's relativity faulty in that respect. Einstein's published response in 1907 did not dispute the data; he argued that the rival theories' foundational assumptions were not suggested by any wider system of ideas and so had "rather small probability." That is a theorist declining to be moved by the best available experiment, and it is only a virtue in hindsight. In 1908 Bucherer ran a more precise experiment on beta rays from radium, got the opposite answer, and wrote to Einstein on 7 September 1908: "First of all I would like to take the liberty of informing you that I have proved the validity of the relativity principle beyond any doubt by means of careful experiments."

Planck is why relativity was read at all. As Annalen's theory editor and the most authoritative physicist in Germany, he picked the June paper up at once, lectured on it in Berlin, and put his students on it. Einstein said so himself in 1913: "The attention that this theory so quickly received from colleagues is surely to be ascribed in large part to the resoluteness and warmth with which he intervened for this theory." Planck also corrected it — the definition of force Einstein used in §10 was replaced by Planck's, and Einstein's printed expression for transverse mass is not the one in any modern textbook. Section 3 grades that.

The light quantum was rejected by nearly everyone for eighteen years. Millikan spent a decade trying to break it and instead measured Planck's constant with it, writing in January 1916 that Einstein's photoelectric equation "cannot in my judgment be looked upon at present as resting upon any sort of a satisfactory theoretical foundation," even though "it actually represents very accurately the behavior" of the effect. The equation was right; the confirmer refused the physics. The clearest statement of the profession's position is the 1913 recommendation by which Planck, Nernst, Rubens and Warburg proposed Einstein for the Prussian Academy: "That he may sometimes have missed the target in his speculations, as, for example, in his hypothesis of light-quanta, cannot really be held too much against him, for it is not possible to introduce really new ideas even in the most exact sciences without sometimes taking a risk." His nominators apologised for the paper he thought was his best. Acceptance arrived with Compton's scattering experiment in 1923; the word photon was coined by Gilbert Lewis in 1926.

Brownian motion was confirmed and settled a different argument. From 1908 Jean Perrin's group in Paris made the measurements Einstein's formula called for, extracted Avogadro's number, and in doing so converted the last serious holdouts against atomism. Wilhelm Ostwald, the most eminent of them, conceded in print. Perrin took the 1926 Nobel Prize for it.

And the career did not move. Einstein applied to habilitate at the University of Bern on 17 June 1907; the faculty rejected the application on 28 October 1907, on the entirely procedural ground that the required habilitation thesis was not in the file. He submitted one, gave a trial lecture on 27 February 1908, and received the venia docendi on 28 February 1908. He remained a patent examiner until 16 July 1909 — four years and one month after the relativity paper was received. The 1921 Nobel Prize, awarded in 1922, was "for his services to Theoretical Physics, and especially for his discovery of the law of the photoelectric effect." The committee, with relativity still contested and under attack from philosophers, went out of its way not to award it for the thing everyone associates with him.

The label came later. "Annus mirabilis" belonged to Newton's 1665–66 long before it belonged to Einstein, and John Stachel's 1998 collection Einstein's Miraculous Year fixed the canonical set at five papers. Which of the five you count, and therefore whether the miracle year contains four papers or five, is a decision made by editors ninety years after the fact.

Why a reading would cite it

Four occasions, in descending order of how well the entry fits. The first two are the ones the proposal named and they are strong. The third is stronger than the proposal knew. The fourth is a caution about time.

First, and it is the reason to keep this file: the lone-genius narrative applied to AI research, cited with the corrective attached. The bare fact is real and it is extraordinary. Between March and September 1905 a government clerk with no laboratory, no research budget, no institutional affiliation and no colleagues in his field sent five papers to one journal, three of which are now foundational, and the total capital expenditure was postage. Nothing about that is exaggerated and a reading is entitled to say it.

Set it beside what this project's own ledgers record about how frontier AI research is currently funded. In the first half of 2026, global venture funding hit $510 billion and OpenAI and Anthropic between them took $217 billion of it — 43% of every venture dollar deployed worldwide, in any sector, in six months. Anthropic's Series H alone was $65 billion at a $965 billion post-money valuation. Alphabet's 2026 capital expenditure guidance is $195–205 billion; Amazon's is $220 billion; Microsoft's net property and equipment went from $205.0 billion to $313.1 billion in a single fiscal year. Stanford's AI Index 2026 — named in this project's concentration ledger as on-lens and not deposited, because it landed four days outside the window — reports that five hyperscalers control more than two-thirds of global AI compute and that over 90% of notable models come from industry. That is the opposite end of a spectrum whose other end is one man's evenings in Bern.

The corrective is the citable part, and it has three pieces. One: he was not outside the system, he was outside a university. He had the training, the journal relationship, five prior publications and a doctorate in progress. The equivalent today is not a hobbyist; it is a competent researcher without a compute allocation, which is a real and different condition. Two: 1905 is the case everyone reaches for precisely because instances are so rare that the same two get cited — Newton's plague years and this. A base rate assembled from the two most celebrated outliers in the history of science is not a base rate. Three, and this is the one that actually bites: the year proves nothing about how cheap discovery is, because the papers were worth nothing until other people's expensive apparatus checked them. Kaufmann's deflection tubes, Bucherer's radium beta-ray rig, Millikan's decade, Perrin's microscopes, Compton's X-ray spectrometer. Section 3 grades each. The lone-genius story is the first half of a two-part structure and citing only the first half inverts its meaning.

Second: the argument that progress requires scale rather than a better idea — and, as the proposal correctly asks, equally the reverse. This entry is the strongest available exhibit on both sides, which is exactly why it is worth holding rather than deploying.

For the better-idea side: not one of the five papers required an instrument that did not exist. The relativity paper contains no new mathematics — the Lorentz transformations were already written down by Lorentz in 1904, and Poincaré presented "Sur la dynamique de l'électron" to the Académie des Sciences on 5 June 1905, three weeks before Einstein's paper was received. What Einstein supplied was not apparatus or algebra but a decision about which of the existing commitments to drop. The live 2026 counterpart in this project's record is OpenAI's account, published 1 August 2026, of an internal Astra model solving ten open problems in mathematics and theoretical computer science — including an explicit construction of a non-sofic group, open since 1999 — for roughly $2,000 of compute, with machine-checkable Lean proofs. Whatever else that is, it is not a scale result.

For the scale side: the confirmations cost far more than the insights, took between three and eighteen years each, and were carried out by institutions. And the sharpest fact on this side is in section 3 — the one paper of the five whose conclusion could be checked against a bench measurement in the short term was wrong, by a factor of 5/2, and stayed wrong in print for five years.

The operational form of the citation is a question, not a verdict: has anybody outside the claimant measured this, and with what? A reading that cites Einstein 1905 to argue that a small team will leapfrog the hyperscalers has used half the entry. A reading that cites it to argue that only capital produces results has used the other half. The entry is worth having because it refuses both. Related: bitter-lesson-2019 and scaling-laws-2020 argue this on the AI side, and maxwell-1865 sets the standard for what a unification has to do to earn the word.

Third, and the proposal did not see it: the dissertation is this canon's cleanest case of a published result that was wrong until an outside laboratory measured it. Einstein derived a viscosity relation, got η = η₀(1 + φ), and published it. Jacques Bancelin, working in Perrin's laboratory, measured η = η₀(1 + 2.9φ). Einstein rechecked and wrote to his student Ludwig Hopf on 27 December 1910: "I have checked my previous calculations and arguments and found no error in them." Hopf found it — a differentiation mistake, a missing factor of 5/2 — and Einstein sent the correction to Perrin on 12 January 1911.

The shape of that is exactly the shape of the discipline this project's readings apply every twelve hours. Compare: BenchJack finding 219 distinct flaws across ten leading agent benchmarks and scoring near-perfect on most without solving a single task; four of the six claimed perfect IMO scores in 2026 never going through official grading, having been scored by a venture capitalist running Claude-based grading agents; the Schema harness self-reporting ~99% on ARC-AGI-3's public set, unverified by ARC Prize. And on the other side, the reason the 2026-08-19 midnight reading treated Anthropic's protein-binder campaign as a finding rather than a claim was that Adaptyv Bio and Twist Bioscience built and measured the designs in their own laboratories. Einstein could not find his own error. Neither can anybody. That is not a fact about Einstein and it is the whole argument for third-party verification — the same argument maxwell-1865 carries at §97, arrived at from the other direction.

Fourth, and it is a caution rather than a case: the clock between a result and its recognition. Relativity was reported refuted in 1906 and vindicated in 1908; the light quantum was published in March 1905, dismissed by its author's own nominators in 1913, confirmed against its confirmer's wishes in 1916, and accepted in 1923. The man who wrote all of it was still examining patents in 1909 and got a Nobel Prize in 1922 for the one his profession had spent eighteen years rejecting, with the citation carefully avoiding the work he is now famous for.

A reading takes the temperature of AI twice a day. That cadence is the project's whole design and it is not in question. What this entry is for is the moment a reading is tempted to say what a result will mean, on the day it lands, when the honest answer is that the field has not had time to check it. Einstein 1905 is the outer bound of how long that check can take when the result is right.

The honest limit, and one disclosure. Nothing in this file is evidence. It is not a capability fact, a compute fact, a policy fact or a market fact; it deposits nothing into history/; it moves nothing. Every 2026 item named above is held with its source in this project's own ledgers and digests and appears here as a citation occasion, not re-argued. This entry is also unusually easy to cite decoratively — "we are waiting for AI's Einstein" costs nothing to write and means nothing — and a reading that reaches for it without either the corrective in occasion one or the question in occasion two is using it as ornament. The test: if the citation survives deleting the name Einstein from it, it is ornament.

The disclosure: the model writing this file is made by one of the two companies whose combined share of world venture funding is quoted in occasion one, and whose protein-design result is used in occasion three as an example of verification done right. Both convenient errors are available — flattering the industry, or flattering this file's own rigour by holding everything to a bar that almost nothing in science clears. The defence is that the standard is stated as a standard with its terms visible, so a later reading that thinks it was applied unevenly can see exactly where.

What it got right, and what it got wrong

Not required for moment. Included because the whole value of this entry is in the gap between what was claimed and what was later found, and because the popular account grades 1905 as five for five, which is not what the record shows. Each paper below carries the date the claim was made, the date it came due, and who supplied the verdict.

Right, and rejected longest — the light quantum. Claim: 18 March 1905, that the energy of light exists as discrete localised quanta absorbed and emitted whole, with the photoelectric consequence spelled out. No due date offered; the technique to test the photoelectric equation properly did not exist. Due date supplied by the world: Millikan's measurements of 1914–16, and Compton's scattering experiment of 1923. Graded: confirmed completely, after eighteen years, over the near-unanimous resistance of the profession, including the resistance of the man who confirmed the equation and of the four men who nominated its author to the Prussian Academy. The useful part of this grade is not that Einstein was right. It is that in 1913 the four most authoritative physicists in Germany, writing a letter of praise, singled this paper out as the example of him missing the target — and they were not being unreasonable, they were reasoning from the evidence available. Correct dissent looks identical to incorrect dissent at the time.

Wrong as published, and corrected by somebody else's laboratory — the dissertation. Claim: submitted 20 July 1905, published AdP 19, 289 (1906), that the viscosity of a suspension goes as η = η₀(1 + φ). Due: immediately, because it was measurable, and Bancelin measured it in Perrin's laboratory and got 2.9 in place of 1. Graded: wrong, by a missing factor of 5/2, found by Ludwig Hopf after Einstein had rechecked his own work and declared it sound in a letter of 27 December 1910, and corrected on 12 January 1911. The corrected treatment yields Avogadro's number as 6.56 × 10²³ — which agreed well with Perrin's independent value and is still about 9% above the modern 6.022 × 10²³, a figure worth stating because the usual telling of this correction stops at "excellent agreement." Two further facts belong with the grade: the corrected paper became the second most cited thing Einstein wrote, and the thesis itself is the most cited, entirely on the strength of industrial applications that have nothing to do with why anyone remembers 1905.

Right, and it settled an argument outside its own subject — Brownian motion. Claim: 11 May 1905, that suspended particles of order 1/1000 mm must show an observable random motion with a calculable mean squared displacement. Due: Perrin's experiments from 1908, published through 1909. Graded: confirmed, and it did more than it claimed — it converted the reality of atoms from a disputed hypothesis into a measured fact, and brought Ostwald across. Perrin's 1926 Nobel Prize is for the confirmation, not the prediction.

Right, and the leading experiment said otherwise for two years — special relativity. Claim: 30 June 1905, two postulates and everything that follows, including the superfluity of the ether. Due: contested on arrival. Kaufmann published in 1906 that the Lorentz–Einstein prediction lay outside his experimental error while two rival models lay inside it, and concluded relativity was faulty. Bucherer's 1908 beta-ray experiment reversed that, and by 1914 Neumann's work had settled it. Graded: right, and the two-year interval in which the best available measurement contradicted it is the most instructive thing about it. Note carefully what Einstein did in 1907: he did not challenge Kaufmann's numbers, he argued that the rival theories were unmotivated. That reasoning was vindicated and it is not a method anyone should generalise from — the identical move made by someone who happened to be wrong would be indistinguishable, and unfalsifiable, and this canon should not be cited to license it.

Right in result, and its 1905 derivation is not the one anybody uses — mass-energy equivalence. Claim: 27 September 1905, that a body emitting energy L as radiation loses mass L/c², generalised to the mass of a body being a measure of its energy content. Due: proposed by Einstein himself against radium salts, and made unmistakable by nuclear physics from the 1930s. Graded: right, and the argument was restricted. The 1905 derivation treats a specific case to first order; Planck gave a general treatment in 1907, Einstein returned to the problem repeatedly, and there is a long literature — Ives in 1952 is the usual citation — arguing that the original reasoning is circular. That critique is contested and this file does not adjudicate it; what is not contested is that the standard modern derivation is not Einstein's 1905 one.

Wrong in the printed formula, for a reason that is itself disputed — transverse mass. In §10 of the relativity paper Einstein gives longitudinal mass as m/(√(1−v²/c²))³ and transverse mass as m/(1−v²/c²). The second is not the expression in any modern textbook, which has m/√(1−v²/c²). The cause is Einstein's definition of force, which Planck replaced. Whether this counts as an error or as a superseded convention is genuinely argued in the literature and I decline to settle it here; what is beyond dispute is that the printed value was abandoned, that Planck supplied the reason, and that Einstein wrote a follow-up in 1906 on determining the ratio of the two masses. The standard 1923 English collection of these papers carries a note at that point — "The definition of force here given is not advantageous, as was first shown by M. Planck" — and I could not establish whether that note is Einstein's or an editor's, so it is quoted here as a note in the translation and not as Einstein's own words.

Wrong about the thing he was proudest of, by his own final grade. Einstein spent the rest of his life unable to accept what quantum mechanics did with the door he had opened, and he graded himself in a letter to Besso on 12 December 1951: "All the fifty years of conscious brooding have brought me no closer to the answer to the question: What are light quanta? Of course today every rascal thinks he knows the answer, but he is deluding himself." Graded: the self-assessment is bleaker than the record justifies, and it is the only self-grade in this file, which is why the entry is not a prediction. One grader is not a grade.

Commonly misused as

Not required for moment. Required in substance, because maxwell-1865 routed the light-speed limit to this file, and because the standard account of 1905 is wrong in several checkable ways that a canon entry repeating them would make worse.

• "Einstein was a patent clerk with no training who overturned physics from nowhere." The most-repeated version and the one that inverts the entry. He held a physics diploma from the ETH Zurich, had five papers in Annalen der Physik before 1905, submitted an accepted doctoral thesis that same year, and was reading current literature with a study circle. What he lacked was a post, a laboratory, a library and colleagues — which is a real deprivation and a different one. The correct short form is that he was a trained, published, unemployed-in-physics researcher with a full-time job, and the outsider story is doing the work of the word "unaffiliated" while borrowing the glamour of "untrained."

• The forged rejection letter, which deserves its own line because it is the narrative manufacturing its own evidence. A letter dated 6 June 1907 circulates widely, in which a "Prof. Wilhelm Heinrich," dean at Bern, rejects Einstein's habilitation as "radical" and "more art than physics." It is a fabrication. Niklaus Bütikofer, the University of Bern's archivist, identified it in May 2016 as "a relatively crude forgery. Not much effort was put into the research" — Wilhelm Heinrich never existed there, the faculties had not yet been separated in 1907, the correspondence would have been in German rather than English, the stamp resembles Hungary's coat of arms, and the street address did not exist until 1931. What actually happened: Einstein applied on 17 June 1907, the faculty declined on 28 October 1907 because "the habilitation thesis required by section 2 of the regulation on habilitation" was not in his file, Dean Gustav Tobler invited him to resubmit, he did, and he received the venia docendi on 28 February 1908. It was a paperwork rejection. A reading citing establishment resistance to a correct idea has a real case available here — the light quantum, 1905 to 1923, documented in section 3 — and does not need the forgery, which is the most-shared piece of evidence for a claim that is true for other reasons.

• "One man, one year, no equipment — so AI research does not need scale." The strongest and most tempting misuse, because the premise is accurate. It fails on the second half of the record: the papers were not knowledge until Kaufmann, Bucherer, Millikan, Perrin and Compton spent years and instruments on them, and one of the five was wrong until a laboratory in Paris measured it. It also fails on selection — the reason this year is famous is that it is nearly unique. Cite the entry for the question who has checked this, and with what?, which it answers well. Do not cite it for a prediction about who will build the next frontier model.

• "Nothing can travel faster than light, and Einstein 1905 proved it." This is the limit maxwell-1865 deferred here, so it gets stated carefully. What the 1905 paper establishes: c is the same in all inertial frames, the Lorentz transformations follow, and within the theory no material body can be accelerated to or beyond c and no signal can be propagated faster. What it is repeatedly made to say: (a) that nothing whatever — no correlation, no phase, no coordinate separation — can exceed c, which is false in three well-understood ways, since phase velocities routinely exceed c without carrying information, quantum entanglement produces correlations at spacelike separation while permitting no signalling, and the recession of distant galaxies is not motion through space; (b) that Einstein proved a metaphysical impossibility rather than derived a consequence of two postulates, which are themselves empirical claims that were doubted, tested, and once reported refuted; (c) that the bound settles any particular engineering argument by itself, when what it actually supplies is a floor on latency — the use maxwell-1865 makes of it in section 2, where fibre latency shapes how far apart two AI datacentres can usefully be. The canon's standing example of a misused limit is godel-incompleteness-1931, conscripted to prove minds are not machines. The error available here is the same shape: promoting a precise result about signals and massive bodies into a universal prohibition on everything.

• "Einstein worked alone." He named Besso in the paper. He read weekly with Solovine and Habicht. He was married to a physicist who had sat the same ETH examinations. Whether Mileva Marić contributed to the physics is genuinely disputed among historians, the documentary basis is thin in both directions, and this file does not resolve it and should not be cited as resolving it. What survives either verdict is that "alone" is wrong on the most conservative reading available, and a reading using 1905 as an argument for solo work is choosing the one adjective the record does not support.

• "Michelson–Morley motivated special relativity." Standard in textbooks and not what the paper says. Einstein referred only to "the unsuccessful attempts to discover any motion of the earth relatively to the 'light medium'" — no experiment named, no citation, and the historical question of how much weight Michelson–Morley carried for him is argued to this day. The paper's stated motivation is the magnet and the conductor: a theoretical asymmetry with no experimental anomaly attached at all. That distinction is worth preserving whenever someone claims a theory was forced by an experimental result.

• "1905 changed physics overnight." October 1907: habilitation rejected. July 1909: still a patent examiner. 1906: leading experiment reports relativity refuted. 1913: his nominators to the Prussian Academy name his own favourite paper as the example of him missing the target. 1922: Nobel Prize awarded, for the photoelectric law, with the committee steering deliberately clear of relativity. A reading tempted to grade a 2026 result on the week it lands should read that sequence first.

• "Four papers." Five, and the four he promised Habicht in May are not the four now collected under the name — his four included the dissertation and excluded the mass-energy note, which he had not yet thought of. The canonical set is an editorial decision taken decades later, and the discrepancy is a small but exact demonstration that even the most famous story in modern science has had its boundaries drawn by someone with a publishing deadline.

• "Fiction and history settle it." Nothing in this file is evidence, nothing here is deposited in the ledger, and nothing here touches the needle. The 2026 items in section 2 are real and are held with their sources in this project's own ledgers and digests; they appear here as citation occasions and are not re-argued. No needle, no score and no landmark is placed by this file.

Sources

Primary, read via transcription. On the Electrodynamics of Moving Bodies — every quotation above (the magnet-and-conductor opening, the "unsuccessful attempts" clause, both postulates, the "superfluous" ether, the Besso acknowledgment, and the §10 mass expressions) is from the Perrett and Jeffery translation as hosted at Fourmilab. Does the Inertia of a Body Depend Upon Its Energy Content? — the L/c² statement, the generalisation, and the radium-salts remark are from the same translation at Fourmilab. This is a weaker link than it looks and is flagged as one: these are the 1923 Principle of Relativity translations, not the German originals, and the note attributing the force-definition improvement to Planck may be an editor's rather than Einstein's — I could not establish which, and have said so in section 3 rather than resolve it silently. The light-quantum paper's opening sentence, the statement of the quantum hypothesis and the photoelectric result are quoted from an English rendering at chem.academy, which also carries a self-contradictory publication date and is the weakest source carrying a quotation in this file; the wording is consistent with every secondary account below, which is corroboration and not verification. I did not read any of the German originals. The Wikisource translation of the light-quantum paper returned HTTP 404 at the URL I tried.

Bibliographic detail. Volume, page and title data for all Einstein papers in Annalen der Physik from 1901 to 1906, including the 1906 publication of the dissertation at AdP 19, 289 and the 1906 follow-ups on Brownian motion and on light emission and absorption, are from the journal's own historic-papers listing hosted at Universität Augsburg. Dates of receipt and publication are from Wikipedia, "Annus mirabilis papers" — tertiary, used for dates only, and it gives 18 March for the light-quantum paper where some sources give 17 March. The Library of Congress research guide to the 1905 papers, which would have been the right source for this, returned HTTP 403 and was not read.

The dissertation and its correction. Norbert Straumann, "On Einstein's Doctoral Thesis," arXiv:physics/0504201, read via extract — source for the 20 July 1905 submission, Kleiner returning it as too short, Bancelin's η₀(1 + 2.9φ) measurement, Einstein's 27 December 1910 letter to Hopf, the missing factor of 5/2, the corrected Avogadro number of 6.56 × 10²³, the 12 January 1911 communication to Perrin, and the claim that the thesis is Einstein's most-cited paper and the 1911 correction his second most cited. I did not verify the degree-conferral date and have not stated one.

Read via search results rather than in the original, and flagged as relayed. The Habicht letter of May 1905 (dated 18 or 25 May by different accounts) is quoted from secondary transcriptions and I did not consult the Collected Papers. The Planck, Nernst, Rubens and Warburg recommendation of 1913 is likewise relayed. Kaufmann's 1906 conclusion, Bucherer's 1908 reversal and his 7 September 1908 letter to Einstein come from the Wikipedia account of the Kaufmann–Bucherer–Neumann experiments and associated Wikisource translations. Millikan's January 1916 assessment is relayed from APS Physics centennial coverage; the much-quoted Millikan line about spending ten years testing Einstein's equation could not be sourced to a primary and is deliberately not used above. Perrin's confirmation, Ostwald's concession and the 1926 Nobel Prize are from standard reference accounts. The Poincaré note "Sur la dynamique de l'électron," presented 5 June 1905, Comptes rendus 140, 1504–1508, is cited from secondary literature on Poincaré and relativity and was not read.

Employment and career. Start date of 23 June 1902, level III technical expert, 3,500 francs, promotion to level II on 10 March 1906 at 4,500 francs, and resignation on 16 July 1909, from the Swiss Federal Institute of Intellectual Property's own FAQ — the successor institution to his employer, and the strongest source in this file. The 1907 habilitation sequence and the exposure of the forged rejection letter are from the University of Bern's own account, including Bütikofer's May 2016 identification, his "relatively crude forgery" assessment, the five internal errors, the 28 October 1907 faculty decision quoting the habilitation regulation, Dean Gustav Tobler's invitation to resubmit, and the venia docendi of 28 February 1908.

The Nobel Prize. The citation wording — "for his services to Theoretical Physics, and especially for his discovery of the law of the photoelectric effect" — and the 1921 award being presented in 1922 by Arrhenius are from NobelPrize.org. That the committee deliberately avoided relativity is well attested across the secondary literature; the widely quoted Arrhenius telegram wording about the prize being given "without taking into account the value which will be accorded your relativity and gravitation theories" is relayed and was not verified against a primary, so it is not quoted above.

Naming and framing. That "annus mirabilis" attached to Newton's 1665–66 long before 1905, and that John Stachel's 1998 Einstein's Miraculous Year fixed the canonical set at five papers, are from Wikipedia and Wikibooks; the specific claim that Stachel first applied the label to 1905 rests on a tertiary source and is stated above only as the collection fixing the set, not as a coinage. The Besso letter of 12 December 1951 is quoted from a widely reproduced transcription; the original is in the Einstein Archives and I did not read it. That the relativity paper carries no citations and names only Newton, Maxwell, Hertz, Doppler and Lorentz is corroborated across several secondary discussions and is checkable against the Fourmilab text above.

The 2026 and present-day material. Every AI fact in section 2 is taken from this project's own records and is named there as a citation occasion only. The $510 billion global venture total and the $217 billion taken by OpenAI and Anthropic, the $65 billion Series H at a $965 billion valuation, and the Stanford AI Index 2026 figures on hyperscaler compute share and industry model share (recorded there as on-lens and not deposited, having landed four days outside the window) are from history/genesis/concentration.md. The Alphabet, Amazon and Microsoft capital and property figures are from history/genesis/compute-and-infrastructure.md. BenchJack's 219 flaws, the four ungraded IMO scores and the Schema self-report are from history/genesis/capabilities.md. The Astra ten-open-problems result of 1 August 2026 at roughly $2,000 of compute, and the Anthropic protein-binder campaign with wet-lab work by Adaptyv Bio and Twist Bioscience, are from digests/2026-08-19-00.md. Those files hold the primary links. Nothing in this file is deposited as evidence, nothing here is a source list for a reading to reuse, and nothing here places a needle, a score or a landmark.