Blondlot "N rays": review by a PhD mathematician

Source gap (repeated from Step 1). The only document is a Google Books scan of the 1905 English translation of "N" Rays (161 PDF pages) with no text layer. I read the key passages from rendered page images [read: page image] and scratch OCR checked against images. Wood's test, replications and the polarization history of X rays are NOT in the folder; they rest on [web] and [memory], tagged below. Every calculation below is mine, run with numbers printed in the book.

Page references are book page numbers; PDF page = book page + 25 for the body (e.g. table on book p.58 = PDF p.83).

Corrections to Step 1

1. X-ray polarization. facts.json says X rays from a tube are "essentially unpolarized in this manner". That is too strong. Barkla showed in 1904-05 that X rays can be polarized (scattering experiments) [web] https://en.wikipedia.org/wiki/Charles_Glover_Barkla. Bremsstrahlung (the braking radiation from a tube) is partially polarized, with the electric vector mostly in the plane of the electron direction and the photon direction [memory, high]. Blondlot's "plane of action" (plane of the X ray and the cathode ray, book pp.3-6) has exactly that geometry. So his orientation dependence is not absurd as a description of geometry. What fails is the detector and the optical-rotation claims (see Claim 1).
2. "Quartz index ~2.9" in facts.json is attributed to X rays. In the book the 2.93, 2.62, 2.436, 2.29 values are for Auer-burner "N rays" (book pp.14-15), not X rays. The Step 1 statement that this "contradicts X-ray and optical physics" is true but the attribution should read "N rays".
3. Wavelength unit. Step 1 calls it an "inference". It can be pinned down from the text: on book p.59 Blondlot counts how many N-ray wavelengths fit in one yellow-light wavelength and gets 0.0085 vs 0.0081 (gratings). That only makes sense if the unit is the micrometre (yellow = 0.589 micrometre, ratio about 69) [read: page image p.84]. So 0.0081 to 0.0176 micrometre = 8.1 to 17.6 nm, photon energy 152 to 70 eV (E[eV] = 1239.8 / lambda[nm]). Unit = micrometre is [read] by inference, high confidence. A second small table (book p.70) gives N1 rays with lambda 0.003 to 0.0083 and indices 1.004 to 1.041, which Step 1 did not list.
4. Step 1's other facts (table values, 65 cm of copper in note 5, "impressionist painter", Wood's test) I checked against the page images and the Wikipedia N-ray page [web] https://en.wikipedia.org/wiki/N_ray and they are correct. The Wikipedia page also says Kelvin, Crookes, Lummer and Rubens failed to replicate it [web]. Rubens appears in Blondlot's own text as having "made the same observation" for the phosphorescence effect (book p.30, note: "M. H. Rubens made the same observation, as he was kind enough to write me"). I could not check the source of that statement beyond the book itself.

Notation and units used

• n = refractive index; delta = n - 1. A = prism apex angle. D = deviation of a ray.
• lambda = wavelength; 1 nm = 0.001 micrometre. E = photon energy.
• Grating equation: d sin(theta_m) = m lambda, with d = spacing between lines = 1 mm / (lines per mm).

---

Claim 1. A tiny spark gap changes brightness with orientation near an X-ray tube; Blondlot reads this as polarization.

a. Internal consistency. Partly. - The logic is "brightness varies with gap angle, therefore like a Nicol prism, therefore polarized" (book p.4). He gives no law. A real analyzer obeys Malus's law, I(theta) = I0 cos^2(theta), where theta is the angle between the polarization direction and the analyzer axis. Blondlot states only "maximum when horizontal, minimum when vertical, intermediate between" (book pp.3-4). That does not distinguish polarization from any other anisotropy (leads acting as antennas, tube fields, geometry of the 40 cm aluminium sheet, the orientation of the electrodes to the induction-coil field). The control is missing: a gap rotated with no tube running. - Hidden assumption: the spark gap is a linear polarization detector with no dependence on its own wires. The leads are "flexible wires" bent to rotate the gap (book p.3), so rotating the gap also moves the leads. - The "rotation" results are unsupported: "Quartz and lump-sugar rotate the plane ... I obtained rotations of 40 degrees" (book p.6, [read: page image p.29]). No thickness, wavelength or direction relative to the optic axis is given. Quartz shows optical rotation only for light travelling along the optic axis, where the linear birefringence cancels; across the axis (as in the March 1903 note, "at right angles to its axis", book p.7) it acts as a birefringent plate [web] https://en.wikipedia.org/wiki/Optical_rotation. A lump of sugar is a polycrystalline aggregate: scattering would depolarize, not rotate. So this is internally inconsistent with the textbook definition of optical activity. - Note 2 (book p.78 note) is a post-hoc rescue: if the true rotation is huge, the observed 18 degrees may be "the remainders after subtracting 360 degrees". Rotation scales roughly as 1/lambda^2; quartz is about 21.7 deg/mm at 589 nm [memory, medium, quartz value not checked on the web page I could read], so at 8.15 nm the same rule gives about 1.1e5 deg/mm. A free choice of the integer k in "40 + 360k" cannot be refuted by any measurement of 40 degrees, so the rescue makes the statement untestable. - Mica claim (book p.8): a single "sheet of mica" at 45 degrees destroys linear polarization, a second crossed sheet restores it. For a real retarder the retardation is 2 pi (delta n) d / lambda. With delta n about 0.04 [memory, medium] and d = 20 micrometres, at lambda = 8.15 nm this is about 98 full waves (computed). A thick sheet at 8 nm on a broadband source averages out any ellipticity; a particular "quarter-wave" behaviour would need a thickness near 50 nm. The text gives no thickness. Note this is a result of the early "X ray" paper, later reassigned to N rays by Blondlot himself.

b. Consistency with experiment. Weak. A spark responding to X rays as such is plausible (X rays ionise gas and affect sparking). Genuine X-ray polarization was first shown by Barkla in 1904-05 using scattering [web] (Barkla link above), after Blondlot's note of February 1903. Nobody has reproduced optical rotation of X rays by quartz or sugar [memory, high]. Blondlot himself concluded in March 1903 that the "X ray" effects were not Roentgen rays (book pp.10-11), which retracts the claim as stated.

c. Tested? Not replicated for X rays. For "N rays" generally, see Claim 2. Blondlot's own retraction (Preliminary Notice, notes 1, 2) says "the experiments are correct, and the rectification only applies to the nature of the rays" (book p.77 note 1, [read: page image]). No independent test of the spark-gap orientation effect is in the folder.

d. If true. Optical rotation of X rays by sugar would require a new interaction at 0.1 nm wavelengths; mainstream X-ray optics (refractive index within 1e-5 of 1, no ordinary lenses or prisms) would be wrong.

Verdict (split). - Weak form (the spark gap is orientation-sensitive near the tube and the geometry follows the bremsstrahlung polarization plane): Plausible, untested as a measurement of X-ray polarization by a spark gap; the later result of Barkla supports the underlying physics, not Blondlot's detector. - Strong form (X rays are rotated by quartz/sugar; mica gives elliptic polarization): Internally inconsistent (rotation across the quartz axis, sugar lump), and the 360-degree rescue makes it Not testable as stated.

---

Claim 2. Ordinary sources emit "N rays" that pass through metals, wood and black paper and are stopped by water and wet paper.

a. Internal consistency. No quantitative law anywhere. Transmission is binary ("traverses" / "does not"). A genuine transmission measurement would give an attenuation law I = I0 exp(-mu x). The list on book p.17 mixes thicknesses: copper 0.2 mm passes, lead 4 mm "arrests", platinum 4 mm arrests, aluminium 1.5 cm passes (book p.50), later copper 65 cm passes (note 5), and later still lead passes too ("opacity of lead is due to oxide and carbonate", note 5). A property that changes classification with each new experiment cannot be checked. Further ad hoc adjustments: - Rock salt 3 mm first "opaque" (book p.16), then "transparent" once polished (note 4). Roughened mica "arrests", polished mica "lets through" (book p.60 and note 3): surfaces must be optical quality, yet metals like copper "pass N rays" at 65 cm. - Water arrests the rays, and the eye is 98.6 % water. Blondlot concludes that salt must make water transparent, tests it, and finds salt water transparent (book p.45). A "very small quantity of sodium chloride" suffices. This was constructed to fit the eye observation. - The cold platinum sheet is opaque but hot platinum is transparent (book p.33). - An N-ray "focus" by a quartz lens (book p.14). Check of the lens formula, using his numbers: lens f = 12 cm for yellow (n_yellow about 1.544 for quartz [memory, high]), object distance p = 26.5 cm, image distance p' = 13.9 cm [read: page image p.37]. Thin lens: 1/f' = 1/p + 1/p' gives f' = 9.12 cm. The lensmaker rule f' = f (n_yellow - 1)/(n' - 1) then gives n' = 1.72, not 2.93. For n' = 2.93 the image would be at p' = 3.9 cm (and 4.75, 5.5, 6.3 cm for 2.62, 2.436, 2.29). I cannot reproduce the quoted index from the printed numbers. Either p' is not the quantity I took it to be (possible: only "about" 13.9, and "the lens formula" is not written out), or the index was not computed from these data. Flag: unreconciled, not proven wrong.

b. Consistency with experiment. Strongly inconsistent, if "radiation" means electromagnetic radiation of 8 to 18 nm: - Photon energy 70 to 152 eV (computed above). Aluminium, copper, wood and 65 cm of copper are not transparent to extreme-ultraviolet photons: attenuation lengths in solids at these energies are fractions of a micrometre [memory, medium; I tried the CXRO page https://henke.lbl.gov/optical_constants/atten2.html but it is an interactive form and returned no numbers, so this check failed]. Even a 200 nm length against a 1.5 cm aluminium plate gives a suppression exp(-75000). - Thermal sources cannot emit them: for an Auer mantle at about 2000 K the Boltzmann-type factor exp(-E/kT) at 70 eV is about 10^-176 (computed with k = 8.617e-5 eV/K). For a dull-red plate (900 K) it is 10^-392; for the sun's surface (5800 K) 10^-61. Steel at room temperature: 10^-1176. - Blondlot himself notes the contradiction with Schumann's ultraviolet rays "which are largely absorbed by air; N rays are not" (book p.61) and posits "absorption bands" between them, without evidence. - Wood's test (1904): the prism was removed and a file replaced by wood, the observers still reported the effect [web] https://en.wikipedia.org/wiki/N_ray.

c. Tested? Yes, by others. Lord Kelvin, Crookes, Lummer and Rubens failed to replicate [web] (same URL). Wood's blind-substitution test in Blondlot's own laboratory: effect reported with the essential prism removed [web]. About 120 scientists published about 300 papers in the period, mostly French [web]. The effect was later shown to be physiological, not a change in illumination [web]. I did not read the primary Nature paper. Blondlot's own instrumental checks (thermopile, resistance) found nothing (book pp.29-32): see Claim 5.

d. If true. A radiation of 70 to 150 eV traversing metres of wood, 65 cm of copper and metals would contradict the measured photoabsorption cross-sections and the Planck law for thermal sources, plus the whole structure of atomic photoabsorption. It would also require emission from a non-thermal source.

Verdict. Tested, not supported (failed replications and a blinded-style substitution test, [web]); additionally, as stated with 8-18 nm wavelengths, Refuted by calculation (energy per photon vs. thermal emission and absorption by solids).

---

Claim 3. N rays are refracted, reflected, diffused, dispersed (aluminium-prism indices 1.04 to 1.85) and diffracted.

a. Internal consistency. The arithmetic is largely sound; the physics of the dispersion is not. - Prism geometry checks (my computation): prism A = 27 deg 15 min, normal incidence on the first face, deviation D = arcsin(n sin A) - A. Results: n = 1.04, 1.19, 1.29, 1.36, 1.40, 1.48, 1.68, 1.85 give D = 1.2, 5.8, 9.0, 11.3, 12.6, 15.4, 23.0, 30.6 degrees. The eight "pencils" are therefore spread over 29 degrees. - Verification on book p.56: 60 degree prism, n = 1.42, minimum deviation I compute 30 deg 28 min at incidence 45 deg 14 min (he gives 30 deg 38 min at 45 deg 19 min; the small gap is in the fourth significant figure). For n = 1.67 at the same incidence I get D = 57 deg 43 min, matching his 57 deg 42 min. So the printed calculations are correct. - Lens table (book p.55): for plano-convex lens R = 6.63 cm, a real image requires f < p, i.e. n - 1 > R/p: n > 1.166 for p = 40, 1.221 for p = 30, 1.301 for p = 22. The table's blanks follow this rule exactly: n = 1.04 has no lens values; n = 1.19 has a value only at p = 40; n = 1.29 is blank at p = 22. This is a genuine consistency in the geometry. It is also what an observer who knows which index he is looking for and where the image "should" be would produce. - The lens values for the same ray drift: for the claimed prism index 1.29 the lens table says 1.36 (p=40), 1.31 (p=30); for 1.19 it says 1.20; for 1.85 it says 1.86, 1.91, 1.91 (book p.55). Differences up to 6.5 %, larger than the "relative error under 4 per cent" he states for the wavelengths. The lens has aperture 6.8 cm and focal length about 7.8 cm for n = 1.85 (f = R/(n-1)): roughly f/1.1, where spherical aberration is large. He does not discuss it. - Wavelength table (book p.58, [read: page image p.83]): the three gratings should give the same lambda. Spreads (my computation of half-range / mean): n = 1.04: values 0.00813, 0.00795, 0.00839, half-range 2.7 %; n = 1.19: 0.0093, 0.0102, 0.0106, half-range 6.5 %; n = 1.85: 0.0176, 0.0171, 0.0184, half-range 3.7 %. The claimed "relative errors do not exceed 4 per cent" (book p.60) is violated by the n = 1.19 row on his own numbers. The "probable value" 0.0099 is also below the mean of the three, 0.0100 (n=1.19), and the n=1.04 and n=1.85 rows round to the stated values, so the table is otherwise internally tidy. - Grating geometry: 200 lines/mm gives d = 5 micrometres. For lambda = 8.15 nm, the tenth order sits at 0.93 degrees from the axis (1.87 degrees between tenth order left and right); for 17.6 nm, 2.0 degrees (4.0 degrees between). A grating of 50 lines/mm at 8.15 nm gives 0.47 degrees between tenth orders. These are measurable angles with a mirror and telescope as he says; so the method is self-consistent on paper. - Newton's rings (book p.59): counting about 69 rings between two yellow dark rings (0.589/0.0085) with a "very wide" ring pattern and a bare eye on a phosphorescent slit. The 5 % agreement with the gratings is called "valuable control", but the ring-count method is called "inferior" by Blondlot and requires interface roughness well below 8 nm [memory, medium: polished optical surfaces are typically good to a few nm to tens of nm]. I did not check this. - Dispersion sign. Table, book p.58: n = 1.04 at 0.00815, 1.19 at 0.0099, 1.40 at 0.0117, 1.68 at 0.0146, 1.85 at 0.0176 micrometre: n rises with wavelength. Blondlot says so (book p.61: "the wavelength of N rays increases with their refractive index, contrary to what occurs with luminous radiations"). Worse, the second table (book p.70) extends this down to n = 1.004 at 0.003 micrometre. Use the Lorentz model from Griffiths (Introduction to Electrodynamics, Ch. 9, dispersion section): n(w) = 1 + (N q^2 / 2 m eps0) sum_j f_j / (w_j^2 - w^2), valid away from resonance (w_j = resonance angular frequencies, f_j = positive oscillator strengths). In any transparent region dn/dw = sum 2 w f_j / (w_j^2 - w^2)^2 > 0, so n must rise with frequency, i.e. fall with wavelength. This is the normal dispersion. A medium with n rising with wavelength across a factor of 6 in wavelength with no absorption (aluminium prism passing 8 to 18 nm) requires negative oscillator strengths, i.e. gain; passive media cannot do it (Kramers-Kronig argument). Further, far above all resonances the model gives n < 1: n^2 = 1 - wp^2/w^2 with wp the plasma frequency. For aluminium (plasma energy about 15 eV [memory, medium]) the free-electron estimate gives n = 0.977 at 70 eV and 0.995 at 152 eV (my computation). So n = 1.04 to 1.85 for aluminium at these energies is not just unusual, it has the wrong sign. I did not check actual aluminium optical constants (CXRO page failed to give numbers). - Reflection and diffusion by polished/ground glass: a ray model of "light" with no wavelength constraint on surface finish; he treats polish as evidence of a short wavelength (book p.60). This is consistent qualitatively, and also what most radiation does.

b. Consistency with experiment. Inconsistent. At 8 to 18 nm refractive indices of all solids are below 1 (the reason EUV optics use mirrors, not prisms or lenses) [memory, high]; the claimed 1.04-1.85 for aluminium and 2.3-2.9 for quartz are not found. Early in the book Blondlot linked N rays to Rubens' residual rays (Reststrahlen) of rock salt and quartz index about 2.18 (book p.13), which are far-infrared at tens of micrometres; the later 0.008 micrometre values are about 6000 times shorter and he flags this reversal himself (note 7: "I have since found ... much shorter wavelengths", book p.78 note 7).

c. Tested? See Claim 2 (Wood's removal of the prism with unchanged reports; failures to replicate). I found no independent measurement of the dispersion table in what I could read.

d. If true. Basic dispersion theory (the sign of dn/dw in a passive medium) and the optical constants of aluminium and quartz in the EUV would have to change.

Verdict. Internally inconsistent with ordinary dispersion physics (wrong sign of dn/dw for a transparent passive medium, n > 1 where n < 1 is required), the arithmetic being otherwise tidy; and Tested, not supported as an experimental claim (see Claim 2).

---

Claim 4. N rays have wavelengths 0.008 to 0.018 (micrometre), much shorter than visible light.

a. Internal consistency. The calculation from fringe angles to wavelength is sound (checked above). But: - The claim contradicts the book's own statements: the same radiation is emitted by Auer mantles, flames, heated plates, compressed wood and a steel knife (book pp.18-52). Photon energy 70 to 152 eV from a room-temperature knife: elastic strain energy of steel, upper-bound estimate sigma^2/2E with sigma = 1 GPa and E = 200 GPa gives 2.5e6 J/m3, about 2e-4 eV per atom (my computation, steel values [memory, high for E, medium for sigma]). A 100 eV photon would need the elastic energy from about 5e5 atoms to be channelled into one quantum. The total elastic energy of a 20 cm3 blade is about 50 J, which spread over 1200 years is about 1.3 nW; the average power budget is not violated, the energy per quantum is. - Equating "spectrum radiations ... emitted by the same sources" (book p.52) with 8-18 nm is inconsistent with the Boltzmann suppression for every source he used (Claim 2). - The Newton's rings control (book p.59) agrees to within 4 to 5 % but with the same detector (the eye on a phosphorescent slit). - The first assumption of the book (book pp.13, 18, 23) was "large wavelength" radiations akin to Rubens' rays; he reversed himself in January 1904 (note 7).

b. Consistency with experiment. EUV at 8 to 18 nm is absorbed by air within centimetres to metres and by every solid within micrometres [memory, high for air, medium for solids]; Blondlot's own N rays are not absorbed by air (book p.61). No measured EUV wavelength matches transmission through metals.

c. Tested? The wavelength measurement was never independently repeated that I can find; see Claim 2 for the broader failed replications. I did not find a source for this specific measurement.

d. If true. Atomic photoabsorption of metals would need a transparent window at 8-18 nm and thermal emission would have to produce non-Planckian EUV.

Verdict. Refuted by calculation if taken at face value as an electromagnetic wavelength (energy per quantum vs. any source in the book; transparency of metres of wood and metals); the measurement itself is Tested, not supported (Claim 2).

---

Claim 5. N rays are detected as a small increase in glow of a spark or phosphorescent surface (up when seen normally, down when tangential), by the unaided eye under special viewing rules.

a. Internal consistency. - The detector is the eye and the criterion is "I see it". The instructions for observation (book pp.82-83: do not look at the source, "see the source without looking at it", silence, no smoke, "some people never succeed") make a null observation uninformative: failure is the observer's fault by construction. This makes the claim unfalsifiable within the author's own protocol. - The calibration of the spark is circular: the spark must be adjusted "by a methodical process of trial and error ... till it is sensitive to the action of N rays" (book p.66). A spark that does not respond is "insufficiently regulated" (book p.67, the one failure in forty photographic runs). So "responds" defines "properly regulated". - Blondlot himself writes that angular dependence (bright seen normally, dimmer seen tangentially) means that only one observer "placed exactly in front" sees the effect, and "it would be illusory to try to make an audience witness these experiments" (book p.73, [read]). A property that cannot be shown to two observers at once is not publicly checkable. - Thermal null results are self-consistent and instructive. The electrometer-bridge test (book pp.31-32): a platinum wire at dull red, resistance change with temperature 1.004 per kelvin, 15 divisions per kelvin, 1/4 division detectable, so DeltaT < about 1/60 K. Check: for a wire at 900 K viewed at 0.6 micrometre, the Planck exponent x = 14388 / (0.6 x 900) = 26.6, so glow changes by about 26.6 DeltaT/T; DeltaT = 1/60 K gives only a 0.05 % change in glow, while a visibly detectable 5 % rise needs DeltaT = 1.7 K, which the bridge would have shown at 100 times its threshold. So a temperature route is excluded. But the same logic is fatal: the glow of an incandescent body is a function of its temperature and emissivity alone (Planck/Kirchhoff). A glow that rises without a temperature change, as reported for the wire (book pp.29-33), has no thermal mechanism. Rays absorbed by a 1.5 mm^2 wire to raise the glow 5 % would also need about 1.3 mW (my estimate: radiated power at 900 K about 0.18 W for the 15 mm x 0.1 mm wire, black-body upper bound, DeltaP/P = 4 DeltaT/T), more than a filtered lamp can plausibly deliver after aluminium, wood and wet cardboard. So either the effect is not thermal and not radiative-heating (what Blondlot claims) or it is not in the wire. - The photographic test (book pp.62-68) is Blondlot's best design: alternate halves of a plate and a shield, 5 s intervals with a metronome, repeated. The weaknesses: the operator knows which position is which; no density measurement (densitometry) is reported, only "notably more intense" and photogravure reproductions the book admits do not render the originals "in a satisfactory manner" and in which "the streaks and most of the spots do not exist in the original"; 40 runs, 1 "unsuccessful" (no stated criterion for success); several runs vary the source but the sensor in all of them is a spark tuned to be sensitive.

b. Consistency with experiment. The effect is consistent with known eye behaviour: sub-threshold brightness judgments in the dark are unreliable and shift with expectation, fixation point, and the pupil; Wood's blind test found reports unchanged when the supposed source or prism was removed [web] (N_ray page above). The Wikipedia page says later work showed the brightness change to be physiological [web]. I did not locate the specific later papers.

c. Tested? Yes by others, no replication: see Claim 2.

d. If true. A new sensitivity of phosphors and sparks to a radiation that deposits no heat (book pp.29-32 null results) would conflict with energy conservation unless the mechanism were non-thermal photoexcitation (which would need photon energies in the right range and a detectable power).

Verdict. Tested, not supported (no independent replication; blinded-style substitution by Wood did not change reports [web]); the protocol as written (look away, "some never succeed", no objective detector, auto-selection of "sensitive" sparks) makes the visual claim Not testable as stated by the author's own rules. The photographic form is Never properly tested by a neutral party with densitometry and blinding.

---

Claim 6. Bricks, stones, sea water and ox eyes store N rays; compressed or strained bodies emit them indefinitely.

a. Internal consistency. - No decay law is given. Storage times: quartz lens "twenty minutes" (book p.39, scarcely noticeable), Nernst filament "several hours", pebbles "four days without any sensible diminution" (book p.42), a hardened steel knife "more than twelve centuries" (book p.51). A single storing mechanism with decay time from 20 minutes to over 1000 years is not parametrised. No half-lives are reported. - Energy argument. Blondlot proposes the energy comes from "the potential energy corresponding to the strained state of tempered steel" (book p.52). Order of magnitude: a 20 cm3 steel blade holds at most about 50 J of elastic energy (sigma^2/2E, sigma = 1 GPa, E = 200 GPa [memory, medium for sigma]), which would sustain about 1 nW for 1200 years (computed). A nanowatt is not excluded by energy conservation alone. The quantum-energy problem is the real one (Claim 4): about 2e-4 eV of strain energy per atom against 70 to 150 eV per photon. - The claim that emission "originates exclusively from the blade" and that the tang (untempered) emits nothing (book p.51), and "a chisel that is hardened and softened in turn is active when hard" (book p.50), would be easy to test blind with a coded set of chisels; no such test is described. - Gold, lead, platinum, silver, zinc all "store" N rays after exposure (book p.40) while aluminium, wood, paper and paraffin do not; but earlier "a sheet of lead stops N rays" while metals also pass them. The classification is inconsistent with a single transmission property. - An ox's eye "transparent to N rays in all directions" (book p.46) and "water arrests them" would require the eye to be effectively dry of water for N rays; Blondlot resolves it by adding salt (see Claim 2). - The retarded appearance/disappearance of the effect (book p.37, 41) is explained by storage, but the same retardation is a normal property of the dark-adapted eye and of phosphors, so it does not separate hypotheses.

b. Consistency with experiment. Nothing in known physics gives a store-and-release for EUV in stones, brick or salt water lasting days; thermoluminescence and phosphorescence exist but at visible/UV photon energies and decay on known timescales [memory, medium]. A steel knife from the Merovingian era emitting "like a modern knife" is a single, unblinded anecdote (book p.51).

c. Tested? Not independently, as far as the folder or my web check shows. I could not find a source for a controlled test of the "storage" or "indefinite emission" claim specifically; Wood's test was on the prism and the file/wood substitution (the file was "supposed to be giving off N-rays") [web], which does bear on emission by a strained source.

d. If true. A new, long-lived, non-thermal emission channel from strained solids at EUV photon energies would conflict with solid-state physics of stress relaxation.

Verdict (split). - Emission by a hardened tool steel: Tested, not supported (Wood replaced a file reported as an N-ray source by inert wood, reports unchanged [web]); also Refuted by calculation if the quantum energy is taken as 70-150 eV. - "Storage" in brick, stone, sea water, ox eyes and salts: Never properly tested (no controlled blind trial on record in the sources I could read), and not testable as stated without a decay law and an objective detector.

---

Summary table

What the arithmetic teaches (for the primer)

• The numbers in Blondlot's tables are not random. Prism deviations, the minimum-deviation example, lens image-existence thresholds and grating angles all check. Self-consistent geometry does not show the object measured exists; it shows the observer applied the geometry correctly to positions he judged by eye.
• The dispersion-sign argument is the cleanest mathematical refutation and needs only Griffiths-level Lorentz-model dispersion.
• The strongest quantitative inconsistency is between claims 4 and 6: the quantum energy (70-150 eV) against thermal sources and strain energies.

Sources and tags

• [read: page image] pp.14-15 (lens/quartz index), p.29 (rotation 40 degrees), p.55 (lens table), p.58 (wavelength table), p.70 (N1 table); OCR checked for the remaining pages cited.
• [web] https://en.wikipedia.org/wiki/N_ray (Wood, Nature 29 Sep 1904, about 120 scientists / 300 papers, failed replications by Kelvin, Crookes, Lummer, Rubens, physiological explanation).
• [web] https://en.wikipedia.org/wiki/Charles_Glover_Barkla (X-ray polarization 1904-05).
• [web] https://en.wikipedia.org/wiki/Optical_rotation (quartz rotation along optic axis, birefringence cancelled).
• [web, check failed] https://henke.lbl.gov/optical_constants/atten2.html (interactive form, no numbers returned): Al EUV attenuation lengths remain [memory, medium].
• [memory, high]: bremsstrahlung partial polarization; EUV optics use mirrors; Planck/Boltzmann factors. [memory, medium]: quartz about 21.7 deg/mm at 589 nm; mica birefringence about 0.04; Al plasma energy about 15 eV; steel yield stress; roughness of optical surfaces.