Attribution fails in two directions and the second is the quieter one. A man can be credited with work he did not do, and he can be used without being named. Euler is the standing example of both, in the same week, in this desk’s own files.
The 1736 paper is real and the reasoning is Euler’s. The diagram is not. It was supplied later, by other hands, and attributed backwards — and it has been reprinted ever since by people in a position to check the original. That is not a scandal. It is the ordinary rate at which an unverified citation propagates when the claim is pleasing and the source is in Latin.
Which makes the other direction worth counting. Below, the second column is the one that matters: chapters of the published corpus, with index pages and this desk’s own audit narrative classified out.
| Named in this corpus | chapters |
|---|---|
| χ, the Euler characteristic, used as a quantity | 76 |
| Euler, by name | 56 |
| Runge–Kutta / RK4, used to integrate | 16 |
| Euler’s method, which Runge–Kutta refines | 0 |
| The polyhedron formula, where χ comes from | 0 |
| Euler–Lagrange · the Basel problem | 0 |
| Index of a vector field · hairy ball | 0 |
Sixteen chapters run a refinement of a method nobody here has written down. Seventy-six use a number and none had said where it came from. The correction is a sentence in each case, and the reason to print the table is that neither gap announced itself: both were found by asking the question, not by tripping over an error.
And one of them paid immediately. A result published at this desk on 16 September recorded, as a passing curiosity, that three different objects all had Euler characteristic zero and that χ therefore separated none of them. The explanation is on Page 3, it is entirely Euler’s, and it turns the curiosity into a specification. The arithmetic on this desk’s own equations, including a misreading it corrects, is on Page 4.
Euler writes to Christian Goldbach in St Petersburg. He states the thing in words, with no symbols at all:
Two papers followed, both published 1758 — E230 carries the statement, written in 1750; E231 carries the proof, written the year after. In the same letter: “It astonishes me that these general properties of stereometry have not, as far as I know, been noticed by anyone else.”
He was right to be astonished, and the reason is the whole of it. Everything known about polyhedra before 1750 was metric — volume, surface area, angles, inscribability. Nobody had looked at a solid and simply counted.
Squash the cube, shear it, round its corners past recognition — still 2. It is the first quantity in the subject that survives bending the object.
Richeson’s chapter in the same volume records that Euler’s own proof has a flaw in it, found and repaired by others over the following century. The ordinary fate of a first proof, and no mark against the observation.
*.html and *.md at a named commit, entity-aware, with index pages, audit narrative, project scaffolding and the page doing the counting classified out. A gap nobody prints gets rediscovered by everyone who approaches itA count of this kind is worth exactly its method, so here it is. The instrument is tools/corpus_count.py, which shortlists with git grep and then matches against each file’s text with HTML entities unescaped — because this corpus writes accented names as entities, and a plain pattern silently misses them. Poincaré reads 68 files entity-aware and 58 without. The chapters column excludes the counter from the set it counts, which is the only way a number like this means anything.
| Pattern | files | chapters |
|---|---|---|
| χ | 89 | 76 |
| Euler | 64 | 56 |
| zeta | 32 | 29 |
| Runge–Kutta / RK4 | 20 | 16 |
| Euler characteristic | 19 | 17 |
| Euler product | 17 | 14 |
| genus | 11 | 10 |
| Gauss–Bonnet | 9 | 7 |
| totient | 3 | 2 |
| Poincaré–Hopf | 2 | 1 |
| Königsberg · graph theory | 1 · 1 | 1 · 1 |
| polyhedron formula | 0 | 0 |
| index of a vector field | 0 | 0 |
| hairy ball | 0 | 0 |
| Euler’s method | 0 | 0 |
| Euler–Lagrange | 0 | 0 |
| Basel problem | 0 | 0 |
Euler’s method. Sixteen chapters integrate. Every one of them uses fourth-order Runge–Kutta, which is a refinement of the forward step Euler wrote first, and no page states the relation or the order of either. A reader who wants to know why the step size was chosen has nothing to read.
Euler–Lagrange. This corpus is built on a variational picture and a contact form. The equation that turns a stationary functional into a differential equation is the join between them and is named nowhere.
The Basel problem. Thirty-two files use ζ. The first value anyone computed for it is ζ(2) = π²/6, which is also the first place π ever appeared in an answer nobody expected it in. Fourteen files use the Euler product; the road from the product to the value is not written down.
Each of those is a citation gap, not a mathematics gap. Nothing in the corpus is wrong for lacking them; what is missing is the sentence that lets a reader follow the lineage backwards. The rule this desk works to: use a result and name it, or do not use it.
The three objects are quotients of an annulus: the annulus itself, the annulus with its whole boundary collapsed, and the annulus with one boundary circle collapsed. They are the attractor case, the repeller case and the attractor–repeller pair. Their homologies are different. Their Euler characteristics are not.
| Space | Homology | χ |
|---|---|---|
| annulus A — the attractor | (ℤ, ℤ, 0) | 0 |
| A/∂A — the repeller | (0, ℤ, ℤ) | 0 |
| A/L — the pair, together | (0, 0, 0) | 0 |
Read the first two rows again. They differ by a shift of two degrees — the dimension of the unstable direction. And χ is an alternating sum. A shift by an even number of degrees leaves an alternating sum exactly where it was.
Reaching for χ to separate an attractor from a repeller is asking the 1750 question of a 1920s object. The distinction is the same one this desk has been printing all month in another vocabulary: a measure correctly computed of the wrong quantity. Euler’s number is the absolute measure par excellence — it survives every deformation — and that is exactly why it cannot report a stability type, which is differential.
The verification script computes χ from the alternating sum of cell counts, which is Euler’s own method, and independently from the alternating sum of Betti numbers, which requires a Smith normal form over the integers and was not available to him. The two agree on all six spaces tested:
“Enclosed by plane faces” is not decoration. Glue a grid of quadrilaterals into a torus and the count changes, at every grid size tested — 3×3, 4×3, 5×4, 8×6, all zero. So does the cylinder: 25 − 45 + 20 = 0.
Which is where this matters rather than merely being true. The invariant set this desk works on lives on a cylinder. Euler’s number for it is 0, and Page 4 is what that permits.
Not established here: Poincaré–Hopf, or the hairy ball theorem. The script computes χ for particular surfaces and indices for particular fields; the sentence joining them is a reading of two computed facts and not a derivation of one from the other. Stated so that the join is not mistaken for a proof of it.
The index of a closed curve is the number of turns the field direction makes as you go once round it. Computed here by winding — numerical quadrature, not a lookup — on the four textbook cases first, so that a reading on an unfamiliar field can be believed:
Now the desk’s frozen planar field, ṛ = −(r−1)(r²+r−a) with θ̇ = 1 and a = 5.436564, which carries two invariant circles, at r = 1 and at r₂ = 1.884652. The first question is where it vanishes, and the answer corrects something easy to assume:
The field does not vanish on either invariant circle. ṛ does; θ̇ does not; and an invariant circle is not a fixed point. The only zero of the planar field is the origin. Every circle about the origin, at every radius tested from 0.3 to 10, has index exactly +1, and a small loop enclosing nothing has index 0.
That is Strogatz’s Theorem 6.8.2 satisfied on this desk’s own equations: any closed orbit encloses fixed points whose indices sum to +1, and both invariant circles enclose exactly the origin. The theorem is not being invoked as decoration — it is a constraint, and the equations meet it.
A chapter published on 16 September established that the three-dimensional flow admits no compact isolated invariant set at all, because on the invariant helix the height equation reads ż̇ ≡ 1 — the orbit climbs forever and escapes every bounded window. The proof needs nothing but that identity. What it did not say is why the flow has no fixed point anywhere to begin with.
It is Euler’s reason. A surface admits a nowhere-zero vector field only when its Euler characteristic is zero. The cylinder’s is 0 — counted on Page 3. The sphere’s is 2, which is why a hairy ball cannot be combed. This flow being zero-free and its surface being a cylinder are not two facts.
The Mini-Beast, Volume III of the Principia Orthogona, was released on 26 September 2026, the full moon of Bhadrapada. It exists in two versions, and the difference is stated on every page. Version 1 is the book as taught: a committed PDF with its own DOI, 10.5281/zenodo.22947908. Version 2 is a reviewed pre-release, live on the web now; its release is set for spring 2027, together with the Cajueiro workshops.
The review went chapter by chapter with one guide: name the state and the control; say which of C, K, F and U is claimed, supported or skipped; and give each claim a status — established, model, analogy or open. Thirteen chapters gained a Lean gate file this week, each built against Lean 4.32 and Mathlib, with no sorry and nothing beyond the three standard axioms. Among the results: the fold and its hysteresis proved in full, so that the one shape the book keeps finding — in markets, in geology, in the immune system — has a kernel-checked core; two axioms of the older taxonomy refuted as false; and the number 33 shown to play no role in the theorem that had been cited for it. The g-series is now one canonical page, with teaching labels and levels kept apart.
The review also corrected things the first edition got wrong: a human spinal-cord pilot now peer-reviewed and cited as such, with its limits; dog-study figures that did not come from the study; a lunar-returns t-statistic misquoted; the Pirangi cashew’s size (8,500 m², about 2 acres, not two hectares); and claims that a finite number of operator steps produces large cardinals, which ZFC cannot do. A new chapter, Ψ · Latent Capacity, replaces the page that had drifted from its subject, acquired savant syndrome. What remains for the author to decide — a chapter map, one operator set, the chapter count — is listed on the pages where it arises, not settled here.
The eighty-first General Assembly held its high-level week from 18 to 29 September, across the river from this desk; the general debate opened on 22 September (UN News). Two of its meetings turned on measurement, which is this journal’s subject, and both are reported here only as far as the record goes.
Sea level, 24 September. The Assembly adopted a declaration on sea-level rise, stating that its “cumulative, severe and irreversible risks demand stronger international cooperation.” The number behind it comes from an instrument record: satellites since January 1993, which by the end of 2025 put global mean sea level about 11 cm higher than at the start, with the rise accelerating (WMO, from its State of the Global Climate 2025). Thirty-two years on one stream, and the finding is the curvature, not the level.
Pandemics, 25 September. The second high-level meeting on pandemic prevention, preparedness and response. The WHO Pandemic Agreement was adopted in May 2025; its pathogen access and benefit-sharing annex is still being negotiated, with adoption aimed at the World Health Assembly in May 2027. The meeting’s political declaration is to go forward to the Assembly for further consideration (WHO). An agreement adopted and an annex not yet written are two different states, and the record keeps them apart. What the agreement governs is worth stating exactly: the sharing of pathogen samples and sequence data among states, and how the benefits derived from them are distributed. Deliberate misuse is a separate instrument — the Biological Weapons Convention — and nothing adopted this week bears on it.
The database such a system builds has a second use, and it is the one this desk recognises. Comparison over time is what lets anyone say a sequence is unusual at all: a lineage is only anomalous against a baseline, and the baseline is the record. That value depends on coverage and on continuity — the same property that makes thirty-two years of satellite altimetry an instrument and eight months of it an anecdote.
Which is the argument for building it, and the argument the countries asked to supply it have been making for twenty years. In 2007 Indonesia stopped sending H5N1 samples to the WHO collaborating centres, on the stated ground that samples went out and the vaccines made from them did not come back. The response was the Pandemic Influenza Preparedness Framework of 2011, and PABS is the same question asked again about a wider set of pathogens. The point is not that a record is being withheld from anyone. It is that the record is needed, that no one is yet obliged to build it, and that the countries whose samples would fill it do not yet have it available to them. The annex is where those terms get set, which is why an unwritten annex is a live question and not a formality.
Also on the week’s programme: the SDG Moment (18 September); the Secretary-General’s event on climate action and the just transition, and the fortieth anniversary of the Declaration on the Right to Development (23 September); the twenty-fifth anniversary of the Durban Declaration (28 September); and the International Day for the Total Elimination of Nuclear Weapons (29 September).
Conley — a candidate closed rather than left open. No compact isolating neighbourhood contains any part of the invariant helix, because the height equation reads ż̇ ≡ 1 on it and a compact set is escaped in finite time. A tube around it is empty too, with an attained bound. In the frozen family a genuine index exists on each side of the neutral line and the two differ — (ℤ,ℤ,0) above, (0,ℤ,ℤ) below. And a rescaling that moves the transverse multiplier over sixty orders of magnitude leaves the isolating block conditions fixed, so that multiplier cannot be an index of this kind at all. Block [8] compares the published contact normal form against the system as separately stated: identical through first order in the transverse coordinate, ż̇ = ω on the helix in both. Neither source had printed the comparison.
Smale — a restriction on the whole family, not one system. For ṛ = f(r) with θ̇ constant, the return map is the time-T map of a scalar autonomous equation, so dP/dr₀ = exp ∫f′ — an exponential, hence strictly positive, for every f. The map is monotone, its lap number is 1 at every iterate, and its topological entropy is exactly zero. No horseshoe is available anywhere in the family. The logistic map at μ = 4 is carried as the positive control, reading log 2 at every iterate.
Gelfand — the object identified, and then built. The operator chain is a composition monoid and not an algebra; the corpus’s own page states the fold operator to be nonlinear, and a nonlinear map is not an element of a vector space. Linearised where the commutation argument already lives — a gate and a shift on n sites — there is a genuine C*-algebra and its structure is complete: dim C*(K,S) = n·p, isomorphic to (Mp)⊕n/p with commutant of dimension n/p, where p is the period of the gate mask. Exhaustive over all 248 masks for n = 3…7, exact over the rationals, no violation. Volume I’s non-commutativity theorem becomes the statement p > 1.
Feigin — Virasoro derived instead of quoted. The Witt algebra from vector fields, exact on 2197 integer triples; its second cohomology computed to be one-dimensional on six windows, with the restriction step a raw truncation needs. The degree-zero cocycles are exactly span{m, m³} — m² and m⁵ are checked and are not cocycles — and the coboundaries are exactly the multiples of m, so the −m in m³−m is forced by requiring an uncentred Möbius sl₂, while the 12 is a unit. Dual Coxeter numbers, root counts and dim g for fifteen simple types from Cartan matrices alone, every value agreeing. E₈: h∨ = 30, 240 roots, dim 248, self-dual, critical level −30.
Euler — Pages 1, 3 and 4. Newton — a definition names a quantity and fixes its measure, and one force is given three measures on purpose; e−4π is an accelerative measure where an index is an absolute one. Whitehead and Russell — the stroke realises all sixteen binary truth functions by enumeration, and Nicod’s single primitive proposition, as printed in the second-edition introduction, is a tautology on all 32 rows. Nine Pages, No Proofs — Hardy asks Ramanujan for the proofs, 8 February 1913, and what that implies for a citation whose address will not resolve.
Stated plainly because the distinction is the whole value of the claim. Everything above is checked by a companion script in the standard library — exact integer or rational arithmetic where exactness is available, exhaustion where the box is finite and labelled as exhaustion where it is. None of this week’s seven chapters carries a Lean core. The Gelfand structure theorem and the Witt cohomology are the two that could take one, and neither has been written; the index volume cannot have one yet because Mathlib has no K-theory directory, which is a fact about the library and is recorded as such rather than as a difficulty. The corpus’s standing rule is that a rung is not occupied because a volume is named after it.
tools/corpus_count.py. The corpus is HTML and writes accented names as character entities, which no plain pattern matches. Liénard reads 5 files entity-aware and 2 without; Poincaré 68 and 58; Gödel 27 and 23. Sixty-eight tracked files carry an accented entity.
tools/self_reference.py. The vicious-circle principle as a mechanical test: for each verification script, what set does it count over, and is the script inside it? Over 57 scripts — 0 vicious, 8 guarded, 48 narrow. --strict exits non-zero, so it can gate a build.
Euler’s method, Euler–Lagrange, the Basel problem. Page 2 — three citation gaps this desk has the material for and has not written.
The other two index candidates. An asymptotic index as the height grows without bound, and a relative class on the pair. Untouched, and now the only two left.
A Lean core for the two that could take one. Named above.
Agnès had served her for six years, long enough to know which of Marie's silences meant leave me be and which meant stay close, I am frightened and will not say so. This was the second kind, and had been for four days straight, ever since they'd left Versailles at a pace that left no room for explanation.
"You understand what I am asking of you," Marie said, in the cramped room above the chandler's shop where they had waited out the last two nights.
"You are asking me to be seen," Agnès said. "And you are asking me to be someone worth chasing."
"Yes."
Agnès considered this for a moment, the way she considered most things — carefully, without hurry, in a way that had always made Marie feel, in the strange hierarchy of a court where she outranked her own maid in every visible way, that Agnès was in fact the calmer and wiser of the two of them.
"Then I will need a good cloak," Agnès said. "If I am to be a fleeing lady of quality, I should like to be convincing."
The plan, such as it was, relied on a simple asymmetry: a king's men, sent looking for a woman who had vanished from court, would look for a woman. They would ask innkeepers about a lady traveling with a maid, fine hands, an accent trained out of provincial roughness and into something that could survive a royal drawing room. They would not, in the ordinary run of things, look twice at a slight, quiet young clerk boarding a merchant vessel with ink-stained fingers and a satchel of ledgers, traveling in the retinue of an elderly royal commissioner bound for the colonies.
So Agnès took the cloak, and the coach north along the coast road that led away from La Rochelle, toward Nantes, moving just visibly enough, just carelessly enough, to be noticed and remembered by exactly the sort of person who might later be asked. She would not go far. She would stop at an inn two days' ride distant, claim a fever, claim confusion, claim — when it came to it, and it would come to it — that she had lost her mistress somewhere on the road and did not know where the lady had gone. It would hold long enough. It only needed to hold long enough.
Marie, meanwhile, cut her hair shorter than she had ever cut it in her life, bound what needed binding, borrowed a boy's coat two sizes large from the chandler's own son in exchange for coin that bought both the coat and his silence, and walked to the docks alone, in the dark, carrying nothing that looked like it belonged to a woman who had ever set foot in Versailles.
She saw Fermat before he saw her, which surprised her — she had always assumed his eyes missed nothing, that noticing was simply the permanent condition of him. But he was occupied, at the base of the gangplank, in the particular performance he did so well: an old man fussing over trunks and manifests, a commissioner's clerk hovering at his elbow, the whole tedious business of an official departure playing out exactly as tediously as an official departure should.
She fell into step near the back of his small retinue, shoulders down, satchel clutched the way a nervous young clerk might clutch it, and said nothing.
It was not until they were both aboard, until the gangplank had been drawn up and the last shouted business of the harbor had begun to fall away behind them, that he found a reason to stand beside her at the rail, the two of them apparently strangers watching the same disappearing coastline.
"You cut your hair," he said quietly, not looking at her.
"You said patterns were the one thing your work taught you to notice. I thought I'd give you less to notice."
"It's a very good disguise."
"It has to be. I have no lady-in-waiting for two more days, at least, and no name of my own until we're well past anywhere that name could still be caught."
He was quiet for a moment, watching La Rochelle become a smudge and then a line and then nothing at all.
"I miscalculated the timing," he said. "I want you to know that. Six weeks, I told you. Four days is not six weeks."
"No," she agreed. "But we are both on this ship, and no one on it knows either of our real names, and the coastline is gone." She glanced sideways at him, the first real look either of them had risked since the docks. "I find I would rather trust the arithmetic that got us here than the arithmetic that didn't."
He almost smiled at that — she watched him fight it, the particular vanity of a mathematician being told his error didn't matter, and lose, mostly, except at the corner of his mouth.
"Very well," he said. "We'll call it solved, then, and see what the ocean makes of the proof."
Part 6 of 18 · Fiction, drawn from the documented record; what is invented and what is not is set out in the series’ own note on fact and fiction. © 2026 Pablo Nogueira Grossi · G6 LLC.
Rights notice. Pierre et Mademoiselle is not covered by this issue’s CC BY 4.0 licence. The serial and the work it belongs to — a limited series in development — are © 2026 Pablo Nogueira Grossi / G6 LLC, all rights reserved. No licence to reproduce, adapt, translate or perform it is granted here.
Documented. Fermat died at Castres on 12 January 1665, and that is the record this instalment sails away from. The commission he travels under sits inside a real bureaucratic moment: in 1663 the Compagnie de la France équinoxiale was relaunched — the same name as the colony that founded São Luís in 1612 and lost it in 1615 — and in 1664 Colbert created the Compagnie française des Indes occidentales and merged it in. The route out of La Rochelle — to Tortuga and French Hispaniola, down the Guiana coast toward Cayenne, on to Maranhão — was a working French Atlantic corridor in exactly those years; Bertrand d’Ogeron was appointed to govern Tortuga and the French coast of Hispaniola in June 1665.
Invented. Marie, and Agnès. The commission to inspect French interests in Grão-Pará, and the voyage. That the death was arranged at all — and with it everything in this instalment that makes the departure work: the room above the chandler’s shop, the decoy coach on the Nantes road, the cut hair and the borrowed coat. Marie was at Versailles until she left.
Each issue, when it goes out. No list bought, harvested, or lifted from a conference roster. Unsubscribe by replying with the word stop.
tools/corpus_count.py · tools/self_reference.py — entity-aware counting, and the vicious-circle test. Both standard library only.sorry, standard axioms only.