The Analytical Engine

Filed correctly. idea is where canon/proposals.md put it and idea is right, but the entry underneath the label is not the one the proposal describes, and the difference is the most useful thing in this file. So the kind first, briefly, and then the correction at length, because a reading that cites this entry on the strength of the proposal line would cite it wrongly.

idea holds it because what survives of the Analytical Engine is an intellectual object and nothing else. There is no machine. There is a design — store and mill separated, program held on punched cards outside the machine, conditional branching, looping, a microprogram in pegged barrels, automatic printed output — and every one of those is a concept that had to be invented before it could be drawn. The canon's idea entries are contributions of exactly that shape: shannon-chess-1950, dijkstra-1959, bitter-lesson-2019. This is the earliest of them by 113 years.

moment was the real alternative and I declined it twice over. There are two candidate moments — Joseph Clement downing tools in March 1833, and the Treasury killing the project after the meeting of 11 November 1842 — and both are events in the world, which is what moment is for. But an entry built on either would be an entry about Babbage's misfortunes, and the citation occasion this canon actually has is about the design and its non-existence, which is a standing condition rather than a date.

limit would be wrong and is worth naming as wrong, because it is the mistake this entry is most likely to invite. The proposal's citation line — "when the binding constraint on a system is manufacturing and money rather than insight" — reads like a law, and someone will eventually cite this file as though Babbage established one. He established nothing. This is one case, with one set of causes, several of which turn out on inspection not to be the ones the case is famous for. Section 4 takes that apart.

prediction fails the kind's own rules. Babbage forecast constantly and dated nothing, and the one grand claim he made about the future — section 3 quotes it — has no due date in it at all. The world supplied 2002.

descends_from is [], and the emptiness is accurate. Jacquard's loom (1804) is the direct ancestor — Babbage took the punched card from it and said so — and it is not in canon/ and not in proposals.md, so there is no id to name. Babbage's own Difference Engine No. 1 (from 1821) is the machine whose collapse freed him to design this one, and it is likewise absent.

What matters more is the descent running the other way, and it needs saying because the file downstream cannot be edited from here. lovelace-1843 is a direct descendant of this entry, and its header says so in prose: it records that "Babbage's Analytical Engine design ... is the object the Notes are about; there is no id for it and I am not inventing one." There is one now. When lovelace-1843 is next touched by a job entitled to touch it, its descends_from should gain babbage-analytical-engine-1837. The same is true of difference-engine-1990, whose entire alternate history runs on the premise that this machine got built. Neither edit is mine to make, and neither is urgent: build.py links descent both ways from whatever the headers say, so the link is missing rather than wrong.

What it is

In March 1833 Joseph Clement, the master toolmaker who had spent a decade building Babbage's Difference Engine No. 1, downed tools and fired his workmen over a dispute about compensation for moving his workshop. Clement kept the tools, which under the custom of the trade were his. The engine — about 25,000 parts as designed, four tons, eight feet high, a machine for tabulating polynomials by the method of finite differences — was never finished. A working fragment of roughly 2,000 parts, assembled by Clement in 1832, is about one seventh of it, and still works.

Babbage, in Doron Swade's phrase, was "forcibly distanced from the nuts and bolts of mechanical construction." What he did with the enforced idleness is the subject of this entry. Between 1834 and December 1837 he designed a different kind of machine altogether, and in December 1837 he wrote it up in a 53-page manuscript, On the Mathematical Powers of the Calculating Engine. That document is the reason this entry carries the year 1837 rather than 1834. It was not printed in his lifetime; it first appeared in Brian Randell's The Origins of Digital Computers in 1973, 136 years after it was written. Several reputable sites say Babbage "published a paper" in December 1837, and that is loose: he wrote one.

The machine it describes has an architecture rather than a mechanism. The parts that matter:

A store and a mill, separated. Babbage took both words from the textile trade. The store held the numbers; the mill did the arithmetic and sent results back. That separation of memory from processor is the single structural feature that makes the design a computer rather than a calculator, and Swade notes it "foreshadowed von Neumann's scheme that has dominated computer architectures in the electronic era." The capacity grew across the design's life: Menabrea's 1842 account, which reflects the state of the design around Babbage's 1840 Turin lecture, describes "probably at least two hundred" columns with "forty discs" each; by Passages from the Life of a Philosopher (1864) Babbage specifies "a thousand variables, upon each of which any number not having more than fifty figures can be placed" — about 16 kilobytes, in brass, in 1864.

A program, held outside the machine, on cards. Three kinds — operation cards, variable cards, number cards — each with its own reader, the mechanism lifted directly from Jacquard's loom. The cards could be advanced or backed up under machine control, which is how the design gets loops and conditional branching. There is no stored program: instructions live on fanfold pasteboard and are read in.

A microprogram. The mill's internal procedures were held as pegs in rotating drums Babbage called barrels — functionally microcode, a century and a half before the word.

Output that is part of the machine. A printer, a curve plotter, a bell, and the ability to punch results onto cards for later reading. Automatic capture of machine-replicable results was integral to Babbage's conception of a calculating engine, not a peripheral bolted on.

Speed, stated honestly and small. Babbage's own estimates, on the assumption that no moving part exceeds forty feet per minute: "Sixty additions or subtractions may be completed and printed in one minute. One multiplication of two numbers, each of fifty figures, in one minute. One division of a number having 100 places of figures by another of 50 in one minute."

One caution about all of the above, and it comes from Swade rather than from me: "Babbage nowhere used the terminology invoked here to describe logical features. The use of terms such as program, microprogram, pipelining, iterative looping, conditional branching, and processing is a backward projection from the modern computer age and is fraught with the known historiographic hazards of anachronical interpretation." The features are there. The vocabulary is ours, and it flatters.

It was never built, and it was never finished being designed. These are two separate facts and the second is the one that gets lost. Babbage worked on the Analytical Engine on and off for 37 years, until his death in 1871. The Science Museum's archive holds 183 drawings from the first phase alone — September 1834 to 6 March 1849, catalogued by his son in 1889 — including roughly thirty "plan" sheets, each a top view of the whole machine at some stage of its evolution. Plan 25 (August 1840) is the famous one because Babbage had it engraved; Plan 28 (refined 1843–47) is a later and mechanically simpler approach that trades performance for tractability. Tim Robinson's analysis for the Plan 28 project identifies six distinct developmental phases from 1832 to 1871. There is no single complete design. There are dozens of mutually incompatible ones, because Babbage never stopped improving the machine long enough to build any version of it.

He asked the government for nothing for it. The Treasury had spent £17,500 on Difference Engine No. 1 — the Science Museum's comparison is twenty-two new Stephenson locomotives — and in 1842, after Astronomer Royal George Biddell Airy advised that the engines were worthless, Peel's government withdrew. In 1852 Babbage offered the design for Difference Engine No. 2 to Lord Derby, writing that "I have discharged to the utmost limit every implied obligation I originally contracted with the country." The offer was rebuffed. He built no engine of any kind.

The one piece of the Analytical Engine that ever existed in metal was made by his son. Henry Prevost Babbage completed the iron frame plates in 1880 and finished a portion of the mill with printing apparatus in 1906, and in 1910 published a table of the 1st to 44th multiples of π to twenty-nine places. It contained errors — an incorrect value of π entered at the start, and printing faults at 8π, 9π, 11π and 22π traced to weak springs. The device was not programmable and had no store. Henry gave the drawings and the unfinished work to the South Kensington Museum in 1896.

The Analytical Engine has still not been built. As of this writing, 2026-08-19, that is 189 years after the design paper.

Why a reading would cite it

The occasion is narrow, recurring, and currently very live: the gap between a capability that exists as a design and a capability that exists in the world. Four concrete cases from this project's own record, in descending order of how well the entry fits.

First, and strongest: the binding constraint on the 2026 buildout is a manufacturing queue, and nobody's understanding is anywhere near it. The compute-and-infrastructure ledger for 2026-04-17 to 2026-08-15 is saturated on both sides, and the AImageddon half is largely a record of things that cannot be made fast enough. GE Vernova's gas turbine backlog reached 116 GW with reservations being taken for 2031 delivery against a four-to-five-year shipping lag; load is being connected years ahead of the firm generation ordered to serve it. CoWoS advanced packaging has been the single tightest bottleneck in the AI supply chain for two years, and TSMC's answer is a wafer count — 125–130k a month by year end — not an idea. Amazon added $20 billion to its 2026 capex purely for memory prices, buying no additional capacity at all.

Every one of those is a case where the design is finished, the money is present, the demand is proven, and the thing still does not exist, because somebody has to make it. That is the Analytical Engine's situation stated in 2026 terms, and a reading that wants to say so without sounding like it is excusing the industry has, in this entry, the oldest and cleanest instance of the shape.

Second: the workshop with the tools in it and nothing coming out. Tesla's Q2 report showed the Model S and X lines at Fremont torn out and Optimus lines going in with production not yet started, weeks after Musk conceded production would be "extremely slow at first" — another slip in a chain running back to 2022. The robotics ledger sorts that AImageddon on the explicit ground that robots built, ordered or valued are not robots doing anything. Babbage's own sequence was the same in order and in kind: a decade of drawings, a workshop, a toolmaker, and a fragment. When a reading needs to mark the distance between an installed line and a shipped unit, this is the citation, and it is a fair one in both directions — Babbage was not a fraud, and the entry should not be used to imply that anyone in the 2026 case is either. He was a designer of the highest calibre whose machine did not get made.

Third, and the case the entry is really for: a design is not a manufacturing specification, and mistaking one for the other is the recurring error. Swade, who converted Babbage's drawings into something a 1990 factory could quote against, is blunt about what he found. "For all their richness, the drawings are not sufficiently detailed to serve as a specification for manufacture: No information is provided as to choice of materials, methods of manufacture, order of assembly, lubrication, precision, or finish." Beyond the omissions there were "dimensioning inconsistencies where the same parts are shown differently sized in different views; layout errors," redundant assemblies, and "other errors, one of which (in the mechanism for the carriage of tens) is fatal to the correct operation of the engine." His summary of the whole discrepancy is the sentence a reading should carry: "The gap, in short, is one of engineering completeness rather than of logical or operational principle."

The sharpest present-day instance of that gap is the attempt to close it. Plan 28, founded by John Graham-Cumming in October 2010 for the express purpose of building the Analytical Engine, with Doron Swade joining in 2011, has as of its February 2025 report completed the survey of Babbage's technical archive and Tim Robinson's 150,000-word analysis of the drawings, resolved the design uncertainties that were blocking implementation decisions, and begun exploratory trials with 3D-printed parts and wood-laminate mockups. Fifteen years in, with modern manufacturing, computer simulation and no funding crisis of Babbage's kind, the project's achievement to date is understanding what Babbage designed. There is still no engine and the report names no completion date. A reading that wants to price the distance between "we have the architecture" and "we have the thing" has, in that project, a controlled experiment running in public.

Fourth, and the one that requires care: capability withheld looks identical to capability unbuilt, from outside, and is the opposite fact. The 2026-08-19 midnight reading records OpenAI holding its largest planned frontier RL run, isolating Astra in restricted environments after determining on 7 August that it could not rule the model out of its own Critical cybersecurity category, and extending monitoring to all tool-using inference of that model at a stated cost of roughly 20% of the compute being monitored. Four days earlier, Anthropic raised its catastrophic-misalignment rating and shelved its strongest internal model. In both cases a capability exists and is not in the world.

That is not Babbage. Babbage could not; these labs would not. The entry is worth citing precisely to keep the two apart, because the observable — a described capability with no public object attached — is the same, and the inference a reader draws from it should be opposite. This is also the only use of this entry where the citation cuts toward the utopia column rather than away from it.

The honest limit. This entry is not evidence and cannot be made into evidence. It is not a capability fact, a compute fact, a policy fact or a market fact; it deposits nothing; it moves nothing. Every 2026 item named above is held with its source in this project's own ledgers and digests and is named here as an occasion, not re-argued. And the entry is a poor citation for the thing it will most often be reached for — "AI is 100 years ahead of its substrate" — because that is a counterfactual and section 4 explains why the Babbage version of it does not survive inspection.

What it got right, and what it got wrong

Not required for idea. Included because the design makes checkable claims about what a machine could be, because one of them came due in 2002 with a verdict attached, and because the entry's popular reputation is built on a cause of failure that the evidence does not support. Claim date throughout: December 1837 for the design, 1864 for Passages, unless stated.

Right, and comprehensively — the architecture. Store separated from mill; program on cards, external to the machine and separate from data; conditional branching; iterative loops; microprogrammed control; automatic printed output. Every one of those reappeared in the 1940s. Due: roughly 1941–49. Graded correct, and correct as a set rather than as a lucky item, which is the part that should impress. Babbage did not guess one feature of the computer; he got the organisation.

Right, and it is the claim with a date and a verdict — the machine would have worked. In Passages Babbage wrote: "If, unwarned by my example, any man shall undertake and shall succeed in really constructing an engine embodying in itself the whole of the executive department of mathematical analysis upon different principles or by simpler mechanical means, I have no fear of leaving my reputation in his charge." Claim made: 1864. No due date given; the world supplied one.

Between May 1985 and March 2002 the Science Museum built Babbage's Difference Engine No. 2 — 8,000 parts, five metric tons, eleven feet long, tabulating seventh-order polynomials to 31 decimal places and printing to 30, one result every six seconds. Swade's conclusion is the grade: "It is fair to conclude that had Babbage built the machine, it would have worked... So we can demonstrably say that Babbage was not an impractical dreamer but a designer of the highest caliber. These assertions could not be made with the same conviction without the empirical confirmation of a working machine." Graded: vindicated, empirically, 138 years after the claim, with the enormous caveat that the machine built was the difference engine and not this one. Section 4 holds that caveat open.

Wrong, and it is the failure the entry exists to record — he could not build it, and neither could anyone else for 189 years. Claim made continuously 1834–1871; due at his death in 1871; still outstanding. The proposal line says he was "defeated by machining tolerance and funding," and that is the received account. On the evidence it is at best a third right, and the composition of the remainder is the useful part.

Money was real and is the least interesting cause. The design was never finished, and that is a cause with a mechanism: six phases across 37 years, thirty-odd whole-machine plans, a designer who by his own 1864 account found "that almost every contrivance in [the first Difference Engine] had been superseded by new and more simple mechanism, which the construction of the Analytical Engine had rendered necessary" — and who responded by redesigning rather than by building. And the drawings, even at their most complete, are not manufacturing specifications. Graded: wrong, and wrong for reasons that are mostly his and are mostly not the famous one.

Wrong, and the counterexample is contemporaneous and cheap. Georg Scheutz, a Stockholm printer, and his son Edvard, working from a magazine description and without access to Babbage's drawings, had a working difference engine with a printing unit in 1843. A second, funded by the Swedish Academy of Sciences, was finished in 1853, exhibited in Paris in 1855, and sold to the Dudley Observatory in Albany in 1857 for about £1,000, where it became the first computing machine to work under a United States government contract. Michael Lindgren's argument from that case, which Swade records without accepting, is that Babbage's insistence on the highest precision achievable was "not only unnecessary" but that "the consequent costs and delays contributed to Babbage's failure." Graded: a working machine of the lesser kind existed in Babbage's lifetime, built by two people with no institutional standing for roughly one seventeenth of what the Treasury had already spent. That is not a full explanation of his failure and it is a fact any citation of the funding story has to carry.

Wrong by omission — no indirect addressing. A card names a column. There is no mechanism for computing which column to name, which is what makes arrays, pointers and general recursion awkward or impossible. Raul Rojas showed in July 2024 that the engine's programming model "can be harnessed in order to simulate indirect addressing, a capability that was not included in the original instruction set," and is explicit that the result is "only of theoretical and historical interest, without any practical implications." Graded: a genuine gap in the instruction set, closeable only by construction and not by Babbage.

Wrong by omission — scale, and this is the omission that matters for 2026. Sixty additions a minute. Decimal fixed point, no floating point. A store of a thousand fifty-digit numbers. Steam power, brass, and a machine the size of a small locomotive. Babbage had a complete architecture and no conception whatsoever of the quantities the architecture would eventually be asked to move. The compute-and-infrastructure lens is a lens about quantities — gigawatts, wafers per month, turbine delivery slots — and none of it has any slot in his thinking, because his machine was the only one there was ever going to be.

A middle case, and it cuts against the entry's own reputation — the design influenced less than is claimed. Howard Aiken read Passages from the Life of a Philosopher and cited Babbage; the Harvard Mark I is the machine most often described as the Analytical Engine finally realised. But Aiken's knowledge appears to have come from the autobiography rather than the drawings, and the Mark I as built had no conditional branching — the feature that separates Babbage's design from a very good calculator. The 1940s did not inherit the Analytical Engine; it re-derived most of it, and on the first pass got less. Graded: the design was a correct answer that the world largely did not use, and that is a different and smaller kind of vindication than "the first computer."

Commonly misused as

Not required for idea. Included because the standard account of this machine is wrong in a specific, checkable way, and because a canon entry that repeated the standard account would be worse than no entry.

• "Victorian engineering could not machine to the required precision." This is the myth, it is load-bearing in almost every popular account, and it does not survive the one experiment that tested it. Swade, appointed curator of computing at the Science Museum in 1985: "I was aware that almost every published account of Babbage's efforts attributed his failure to the limitations of 19th-century mechanical engineering. I was both perplexed and incredulous that, with advances in manufacturing, no one had attempted to realize any of his designs once the supposed limitations of his time had been removed." So the build tested it. Michael Wright and Allan Bromley measured parts of Difference Engine No. 1 and established that Clement had made repeated parts within two-thousandths of an inch; that became what Swade calls "the taboo threshold: An absolute condition was that repeated parts would be made to tolerances no finer than this." Modern computer-controlled manufacture was used, but "no part was made more precisely than was achievable by Babbage himself, although possibly by other means." Materials were checked against X-ray microanalysis of surviving Babbage parts at Imperial College. The machine works. Anyone citing 19th-century precision as the cause is citing an explanation that was tested and failed.

Two things make the myth durable and both are worth knowing. Swade also notes that a tolerance — an interval within which a dimension must fall — is itself an anachronism: until the 1850s there was practically no standardisation, each workshop cut its own screw threads, and parts were made one at a time and hand-fitted until they worked. And the build was not easy: the trial piece alone took four years against a one-year plan, the calculating section's 4,000 parts were quoted at £201,000 and came in at £246,000 in 1990 money, and "some 10 percent of the 210 carry levers required final adjustment by hand." Hard is not the same as impossible, and the myth converts one into the other.

• "The Science Museum built Babbage's computer and it worked." They built Difference Engine No. 2, which is not a computer. It tabulates seventh-order polynomials by finite differences and can do nothing else; it has no store, no program, no branch. The Analytical Engine has never been built and the DE2 result does not transfer to it automatically — Swade himself notes that the Analytical Engine's functions "were vastly more demanding" and its mechanisms required "intricacy and complexity well beyond those for the repeated additions in the Difference Engine." What the 2002 machine licenses is a narrower and still large claim: Babbage's designs were sound engineering and his drawings could be brought to a working machine within his own manufacturing envelope. Whether the Analytical Engine would have worked is open, and Plan 28 is the attempt to find out. This is the same general-purpose-versus-special-purpose confusion that difference-engine-1990 grades as a live 2026 error and that lovelace-1843 credits Note A with first drawing; all three entries should be cited together on it.

• "It was Turing-complete." Repeated everywhere, usually with "a century before Turing" attached, and it is looser than it sounds. The engine has branching and loops and a finite store, which makes it a finite-state machine with a large tape, as every physical computer is. The specific modern result — Rojas, July 2024 — establishes universality only by simulating indirect addressing, which the original instruction set did not have, and only for "a working memory of fixed size," and its author says in the abstract that it is "without any practical implications." The honest statement is that Babbage designed the first machine whose organisation is recognisably that of a general-purpose computer. Turing-completeness is a property established about it in 2024, on terms, by someone doing the work Babbage did not.

• "Babbage designed it in full." He did not, and this is the proposal line I am correcting. There are six developmental phases between 1832 and 1871 and roughly thirty whole-machine plans; the versions are not mutually compatible; and the Plan 28 team's first substantial task, running from 2011 to at least 2025, was deciding which Analytical Engine to build. "Designed in full and never built" is a tidier story than "designed repeatedly, never settled, and therefore never built," and the untidy version is the one with a lesson in it.

• "Computing was set back a hundred years." A counterfactual nobody can price, and the transmission evidence runs against it. The 1837 paper sat unprinted until 1973. Aiken, the clearest case of direct influence, worked from the autobiography and built a machine without conditional branching. Turing does not appear to have needed Babbage to arrive at the universal machine. The Analytical Engine may be better understood as a road not taken than as a delayed arrival, and a reading that wants the "we were nearly there" frame should notice it is asserting something about counterfactual history that historians of computing decline to assert.

• Reading this entry as a limit — "the constraint is always manufacturing and money." It is one case, of one machine, with causes that include a toolmaker's contract dispute, an Astronomer Royal's hostility, a designer who could not stop revising, and drawings that were never a specification. The canon's standing example of a misused limit is Gödel, pressed into service to prove minds are not machines; the error available here is smaller and the same in shape — treating a single historical episode as though it established a rule about where bottlenecks live. If a 2026 reading wants to claim that a particular capability is constrained by fabrication rather than insight, it has to carry that claim on the primary record — the turbine backlog, the wafer count, the line that is installed and not producing — and may cite this entry only for what the claim would mean.

• "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 named 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.

Sources

Primary, read directly. Charles Babbage, Passages from the Life of a Philosopher (Longman, Green, 1864), chapter VIII, in John Walker's Fourmilab transcription. Source for the store's stated capacity ("a thousand variables, upon each of which any number not having more than fifty figures can be placed"), the speed estimates, and the mill/store definitions.

L. F. Menabrea, Sketch of the Analytical Engine, trans. Ada Lovelace, 1843, in the Fourmilab transcription. Source for the earlier design's column and disc counts and the twenty-figure multiplication estimate. lovelace-1843 covers this text and its Notes in full, and is the entry to go to for anything about them; this file uses it only for the machine.

H. P. Babbage, "Babbage's Analytical Engine," Monthly Notices of the Royal Astronomical Society 70 (1910), in the Fourmilab transcription. Source for the partial mill, the table of multiples of π, the printing errors and their cause, and the 1896 donation to South Kensington.

The strongest source in this file, read in full. Doron D. Swade, "The Construction of Charles Babbage's Difference Engine No. 2," IEEE Annals of the History of Computing 27:3 (July–September 2005), 70–88, PDF, read page by page rather than in summary. Every Swade quotation above is from it: the "engineering completeness rather than of logical or operational principle" verdict, the deficiencies in the drawings and the fatal carriage-of-tens error, the taboo threshold of two-thousandths of an inch measured from Clement's parts, the anachronism of "tolerance," the £201,000-to-£246,000 quotation, the four-year trial piece, the 10% of carry levers hand-adjusted, the anachronical-projection warning about modern vocabulary, Babbage's 1852 letter to Lord Derby, and the concluding judgment that had Babbage built the machine it would have worked. The Lindgren argument about Scheutz and the costs of Babbage's precision is relayed by Swade at note 26 and I have not read Lindgren.

Read via fetch-and-extract rather than page by page, which is a weaker link and is flagged as one: the Computer History Museum's Babbage pages and A Modern Sequel; the Science Museum's account of the Difference Engines, source for the £17,500 Treasury total and the locomotive comparison; the Science Museum archive entry for the Analytical Engine drawings, source for the 183 drawings, the date range and H. P. Babbage's 1889 cataloguing; the Plan 28 blog and its February 2025 project report, source for the six design phases, Robinson's 150,000 words, the exploratory trials and the absence of a completion date; and the Computer History Museum's note on the numbered plans, source for Plan 25's August 1840 date and Plan 28's character.

Raul Rojas, "Unconventional Universal Computation in Babbage's Analytical Engine," arXiv:2408.03334, submitted 20 July 2024. The abstract was read verbatim and both quotations above are from it; the body of the paper I could not extract and have not read, so nothing here depends on its construction beyond what the abstract states.

Weaker, single-sourced, or contested — flagged. Allan Bromley, "Charles Babbage's Analytical Engine, 1838," Annals of the History of Computing 4:3 (1982), is the standing technical analysis of the design and I could not read it: the PDF would not extract. It is known here only through its abstract and through Swade's citations of it, and any point in this file that would properly rest on Bromley rests on Swade instead. The publication history of the 1837 manuscript — 53 pages, unprinted in Babbage's lifetime, first published in Randell's The Origins of Digital Computers (Springer, 1973) — is assembled from search results and the Springer chapter listing; the Springer full text is paywalled and I did not read it, and several popular sources say the paper was "published" in 1837, which I judge to be loose usage rather than a competing claim. Airy's "worthless" verdict, the stormy meeting of 11 November 1842 and the subsequent withdrawal are secondary and relayed. The Scheutz figures — 1843 first engine, 1853 second, Paris 1855, Dudley Observatory 1857, about £1,000 — are from museum and encyclopedia sources, not from Lindgren or from primary records. Aiken's reliance on Passages and the Mark I's lack of conditional branching are relayed from secondary accounts and are the kind of claim that should be checked against Cohen's work on Aiken before a reading leans on it; the same goes for the frequently repeated "Babbage's notebooks were rediscovered in 1937," which I could not corroborate and have therefore not asserted — Bromley's serious archival work began in 1979.

The 2026 material. Every dated present-day fact in section 2 — the GE Vernova turbine backlog and 2031 reservations, the CoWoS and TSMC wafer figures, Amazon's memory-driven capex increase, the Fremont Optimus lines, the Astra determination and OpenAI's pause, and Anthropic shelving its strongest internal model — is taken from this project's own ledgers under history/genesis/ and from digests/2026-08-19-00.md, which hold the primary links. They are named here as citation occasions only. 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.