Method & Foundations — the argument, made once

The foundations — the argument, made once

Twenty-four items answered one at a time would inherit the book's own problem: the point drowns in volume, and each chapter gets judged without the others in the room. So the general argument is made here once, and the pages link to it instead of re-arguing it. It substitutes for nothing; every claim is still answered on its merits, page by page.

The shape of the argument

Knowledge here is a mesh, not a chain. Read one at a time, the objections often land: Eratosthenes really does assume parallel rays and a sphere before it measures either, and a single pair of shadow angles genuinely cannot separate a far Sun over a globe from a near one over a plane. We concede that on the page itself.

What the isolation misses is that Eratosthenes was never carrying the Sun's distance alone — not even in 240 BC. A generation earlier Aristarchus had put the Sun at least nineteen times the Moon's distance by a method with no connection to shadow angles: at exactly half Moon the Sun–Moon–Earth angle is a right angle, so the Sun's elongation from the Moon gives the ratio directly. He measured 87° and got 19; the true figure is 89.85° and the true ratio 389, so he was short by a factor of twenty, and it did not matter, because the result that kills a near sun is a ratio and needs no distance at all. Every flat model on offer hangs the two at comparable heights, and none survives a factor of nineteen, let alone 389. The method assumes the Sun lights the Moon, which the catalogue below disputes; it need not be granted here, because the illumination is carried independently by the phase sequence and the terminator — Is the Moon a Ball? Anyone could also check, then as now, that the Sun's angular diameter does not change between horizon and zenith, which a lamp a few thousand miles above a plane cannot manage — that is Q8, an observation rather than an instrument. Since then the same number has come back from trigonometric parallax, stellar aberration and radar ranging of Venus: three unrelated instruments, one answer. Knock over one pillar of fifty and forty-nine are still standing, which is what the ledger below is for.

A measurement that is no longer performed is not a measurement that failed. Q11 asks why no scientist has tested a Compton generator in Antarctica. A Compton generator is a 1913 teaching demonstration good to about one per cent; ring-laser gyroscopes now measure the same rotation continuously to parts in 10⁸, several of them, permanently, at latitudes spread across the world. Nobody levels a kilometre of water for the same reason: better instruments exist. Disuse is evidence about instrument economics, not about the result — with one limit. It only answers when the replacement measures the same quantity and is checkable by someone outside the institution that runs it. Where no replacement exists the old experiment should still be run, and its absence is a fair thing to point at; the polar Foucault pendulum is close to that case, and our own page concedes the evidence there is thin.

There is no model on the other side to check, and that is deliberate rather than evasive: “it is much more effective not to present an alternative model at all” (p. 28), and “falsification is independent of replacement” (p. 187). The principle is correct and we concede it. What it forfeits is that a falsification has to be a falsification of something. The local sun in one chapter, the non-straight light in another and the flat celestial plane in a third are never required to hold at the same time, because no chapter is answerable to the next. That concession has its own limit, set out next: falsification needs no replacement, but a positive reading of the data does owe an account of the measurements already taken.

Three things get called falsification

This is the most important distinction on the page, and both sides get it wrong. Falsification of a model, a single reading that does not fit, and a real effect invoked in the wrong regime are three different events, and only the first is what the word is usually taken to mean.

• A reading that does not fit is the ordinary raw material of measurement — an edge case, an instrument fault, an unmodelled local condition, or the beginning of something. Mercury's perihelion advanced 43″ per century more than Newtonian gravity allowed, and the right response was neither to discard Newton nor to explain the residual away; it was to keep measuring it for sixty years. A discrepancy earns attention. It does not by itself carry a model down.
• A real effect in the wrong regime does the most damage in this debate, because nothing in it is false. Light genuinely can be bent through ninety degrees — in a fibre, in a laboratory. Water genuinely can be deformed against gravity — at Nijmegen, in a 16.5 tesla field. Neither is available at ten thousand feet in open air, or in a warehouse. The question is never can this happen but how much of it is available here, and in which direction. Our Polaris page is the worked instance: rescuing the flat measurement needs starlight bent by 19°, refraction supplies 0.09° at that elevation, and it pushes the opposite way. The mechanism is real, two hundred times too small, and the wrong sign.
• Falsification of a model is when a prediction the model actually makes, in a regime where it claims to hold, measurably fails. That is rare, and it is the thing worth arguing about.

The burden scales with the evidence already in hand — measurements, not the number of people who believe a thing. A result supported by many independent instruments takes more to overturn than one resting on a single measurement, because a replacement has to account for all of them at once. And here is the part usually left out: a new reading of the data owes an explanation of the old data, though not of the old conclusions. It has to say what the earlier measurements were measuring, when they were valid, and what is different now. Those numbers were real events in real instruments; they do not stop having happened. General relativity is the model case, and the strongest example available to a sceptic: it contained Newton, reducing to him exactly in weak fields at low speeds, so every ephemeris computed before 1915 stayed good; it discarded the conclusions — absolute simultaneity, the flat background, gravity as a force — and paid for that by explaining where Newton had been right and why; and it predicted 1.75″ of light deflection at the solar limb, twice the Newtonian value, before the measurement was made. It inherited the corpus and stuck its neck out.

The standing argument: geometry cannot decide, so price the flat branch

Five pages lean on one argument, so it is stated here once. A photograph of something far away over water is read against two unknowns at the same time: how much the surface curves, and how much the air bends the light on the way. Those two are exactly degenerate. A flat plane with a refraction coefficient k′ = k − 1 reproduces every sightline of a globe at k, because both corrections go as D²/2R; we have verified that to machine precision. So no photograph, on its own, separates curvature from refraction, and any page that says otherwise is overclaiming in either direction. That is the concession, and it is made in full.

What separates the two is not geometry but thermodynamics. Air bends light in proportion to how its density changes with height, and there is a ceiling on that: k = 503 (P/T²) (0.0342 + dT/dh), which reaches zero at the autoconvective lapse of −34.2 K/km — the gradient at which a layer of air becomes denser above than below and simply overturns. Ordinary air near k ≈ 0.17 needs a lapse of about −6 K/km. The flat branch of the same photograph, at k′ = −0.83, needs about −171 K/km, five times past overturning; a layer that steep is a chimney, not a lens. So the method on every long-path page is the same three steps: concede that the geometry cannot decide; hand both models the same inputs — the same distances, heights and ephemeris, since the objection that an input “assumes the globe” applies to both branches equally; then price what the flat branch needs the atmosphere to do. The worked exemplar is the summit-to-reflection angle on Mirrored Reflections, where the flat branch needs air past absolute zero halfway up the mountain; the same pricing runs through Sunlight Bends in Air, Celestial Globes Exposed, Chicago & Pontchartrain and The Black Swan, and the site’s Reverse Refraction Solver does the arithmetic for any case a reader brings.

The related rule for things that vanish: perspective and angular resolution shrink and blur an object; they do not amputate it. Bottom-first removal means something is blocking — not necessarily curvature, since a wave or a hill will do, but a blocker all the same. Haze can fade a base before a top, but it fades rather than cuts.

Where his own claims collide

Each row is a pair of positions from the book that pull against each other, and both halves are his. The rows are not all the same kind of collision, so each is labelled. Verbatim rows are two stated positions about the physical world that cannot both hold, and need no globe physics to read. Entailment rows are weaker in kind and marked as such: they close only once a further step is supplied, and the row states the step. Every quotation below has been checked against the prerelease text.

One model has to answer to every chapter at once. That is the constraint the globe is under, and it is the constraint the alternative has not yet been put under. None of this implies bad faith — these are positions colliding, not a person being caught. A book assembled chapter by chapter is exactly where you would expect to find them, and finding them is what a review is for.

Settled once, then cited

The first three are established outside this review, by instruments that do not depend on one another; no page here proves them, they are cited. The fourth the book makes a headline claim, so this review derives it on its own page, and the entry is here so the rest of the site does not do it again.

1. The Sun is about 93 million miles away. Radar ranging of Venus, stellar aberration, and trigonometric parallax — three unrelated techniques, one number. Does not rest on Eratosthenes, before or since. Established outside this review.
2. The Earth turns once per sidereal day. Detected by five mutually unrelated physical mechanisms — a spinning mass (Foucault 1852, and every ship's gyrocompass since), the Sagnac effect on light (Michelson & Gale 1925; ring lasers now at parts per billion), the Coriolis force on a vibrating structure, the Sagnac effect on matter waves, and quantum phase coherence in superfluid helium — with no shared calibration chain, six orders of magnitude of precision between crudest and best, and no disagreement. Established outside this review; our drifters page recovers the period from public NOAA data as a check, not the source.
3. Light through air bends by a measured amount, not an assumed one. Geodesy does not assume straight sightlines; it carries a refraction coefficient, conventionally k ≈ 0.13, and corrects for it. Honest limit: near the ground k is not a constant and has been measured from −4 to +16, which is precisely why long-baseline work uses evacuated pipes. Established outside this review.
4. Rocket thrust exists in vacuum. Derived here, not borrowed — it is Claim #2 and Q1. Rockets Don't Push Against Air argues it from the nozzle outwards and Thrust, Measured in Flight gives the in-flight record, cross-checked by independent methods on the same engine, with the ground-chamber artefact quantified at about one per cent in the direction that flatters the chamber. Every other page cites those two.

Where the argument actually is

One cell per page, in the book's own order and under its own section titles. Blue is prose, and the darker the blue the more words. Sand is a page carrying a picture and no sentence on it. A red outline marks a page that states one of the twenty-four claims.

Method. Word counts come from the text layer of the 500-page PDF (Canva export, 3 August 2026) via pdftotext -layout; the image inventory is pdfimages -list on the same file; words set inside an image are not counted, which is why the pictures are reported alongside. Page numbers are the book’s own folios. They sit a constant 26 from the page numbering of the text dump this map was built from (pdftoppm and pdfimages number the file one higher, so the offset against those is 27), but renders of the prerelease differ in their leading pages, so cite the folio, not the offset. Front matter before p. 7 is excluded. If any of this is measured wrong, name the section and we will correct it here.

Two kinds of entry, and only one of them states anything

The book’s claims come in two grammatical forms, printed in the same lists and counted the same way, and the difference decides whether anyone can do anything with them.

A finding asserts something about the world that can be gone and looked at: “mountain shadows are never cast on the bottom of clouds during sunset”; “46% of all stars actually show negative parallax”; “during a solar eclipse, the shadow cast on the Earth is smaller than the Solar System predicts.” Those can be wrong, which is their merit. Somebody can measure a cloud, count the parallaxes, compute the umbra.

A gesture raises a doubt without saying what is wrong: “the explanation for a sun dog is not satisfactory”; “what evidence 100% proves that the Moon rotates?”; “stars visibly passing through the Moon?”. The formula “is not satisfactory” appears six times in the draft, always without saying which step of the explanation fails, or what would count as satisfying it.

The test used: does the entry assert something about the world, or ask about an explanation? An entry proposing an experiment without running it counts as asking. The lists are short and printed in full in the book; anyone who thinks we have graded one wrongly can re-do this in ten minutes. The same test is applied to the twenty-four rows of the catalogue on the landing page, each of which carries a States or Asks tag: eleven state, thirteen ask. A question counts as asking even when it presupposes an observation, and a section is tagged on its thesis.

The defence, and why it does not cover this

The book anticipates this. Its stated method is that falsification is independent of replacement: you may show a model fails without owing anyone a better one. That is correct, and it is the strongest thing about the book’s method. But it answers a different objection. Declining to supply a replacement is legitimate; declining to supply the falsification is not, because that is the thing being claimed. A falsification names a prediction, names an observation, and shows they conflict. “The explanation is not satisfactory” does none of the three. It is a request for a falsification, not one.

It matters more here than elsewhere because the case is explicitly cumulative — Claim #3 is that “there are too many red flags”, so the argument is the length of the list, and a cumulative case needs every item to carry weight. An open question carries none until somebody answers it, and less than none if the answer goes the other way. A question printed in a book asks every reader to do the work its author did not, which is reasonable once or twice and something else when it is half the list; and it means the answers, when someone does the looking, belong to the record as much as the questions did.

So we did the looking

Two of the six “not satisfactory” entries now have worked answers on this site, and in both the answer went against the objection.

• “The moon-tilt illusion explanation is not satisfactory.” The book supplies the explanation itself, thirty-seven pages earlier at p. 129, and calls it visual geometry — a straight line rendered as a curve by the projection, which p. 145 makes the other side’s defining error to ignore. The angle it disputes has had a closed form since 2014; run on the date of the book’s own Example 1 it returns 45°, which is the figure the original questioner wrote on his own diagram and offered as impossible. The Moon-Tilt Illusion
• “The selenelion… is conveniently explained away with, you guessed it, REFRACTION.” It is refraction, and the amount needed is about 25′ per body — 25.5′ for the Moon, 24.3′ for the Sun — against a textbook horizon value of 34.5′ — less than the standard number, not more. The same page reduces the photograph the book prints as the clincher and finds the model had already predicted its geometry — the coverage to two per cent, the shape of the dark edge to within the bloom on a sixty-six-pixel disc. The Selenelion

That says nothing about the remaining four, which we have not worked and might lose. It says what the exercise costs: each of those took a day, and each began by supplying the specific objection the entry had left out. Where we have done the looking, it has not favoured the case the list is making.

In scope, not yet answered

The book declines to advance a positive model, but the project no longer does. Its launch video carries, and its author endorses as representing his own thinking, a geocentric introduction by another presenter, Mark Knight, that commits to specifics — a near sun on a spiralling 24-hour track at a stated altitude, a self-luminous moon at a stated rate, a star field at a stated altitude turning in 23 h 56 m 4 s. Under this review’s own rule the burden attaches wherever a positive reading is offered, so the endorsement moves it, and the endorsed model’s own admissions come into scope with it. We are working through that material and will cite it by timestamp once each quotation has been checked against the audio.