Blondlot: quality engineering (Step 4b)

Inputs checked: _text/manifest.json, digest/N-rays-00bf93.01-04.md (all four read), facts.json, claims.md, physicist.md, mathematician.md, contrarian.md, arbiter.md, primer.md, example.py. Spot checks were made against _text/N-rays-00bf93.txt (the Step 0 OCR text). There is no deep/ folder, because this project was reviewed before digests and deep reads existed.

Page convention in this file: "PDF p.N" is the scan page, which matches the digests and the === page N === markers. "Book p.N" is the printed page number, which the reviews use.

Verdict: MINOR. Coverage is complete and the verdicts are sound. The defects are stale or incorrect statements in facts.json and claims.md, one open question that the text itself answers, one misreading, a small arbiter/primer number mismatch, and a few items of missed material. None of them changes a central verdict.

---

1. Coverage

• Every manifest document is covered. There is one document, and all of it was digested.
• Step 1 read more than the digests. Before Step 0 existed, Step 1 read the whole body from rendered page images plus scratch OCR. The reviewers then re-read the key pages from images. In effect the whole book had a deep read, though no deep/ file records it. That is acceptable for this project.
• Stale coverage statements (defect, minor). facts.json says text_available: "none" and claims.md opens with "It has no usable text layer". Both are now wrong. Step 0 produced a full OCR text of 200,936 characters, and spot checks show it reads well (Section 4). facts.json also puts the non-substantive pages at "1-12 and 122-161". The book actually ends at PDF p.116, and the catalogue runs from p.118.

2. Shortcuts

• No central claim rests on a 1-3 sentence digest alone. Every claim was argued from page-image readings, cited by book page. The load-bearing pages are the wavelength table (book p.58), the 4 per cent statement (p.60), the 1.5 cm aluminium plate (p.50), copper 65 cm (note 5), the viewing rules (pp.82-83) and the photographic method (pp.62-67). Every one of them was read directly by at least one reviewer or the primer author.
• The book was not sampled: Step 1 says it read the whole body, and the citations range across it.
• Acceptable shortcut, flagged: the arbiter says it "did not re-open the book's page images" and relied on the reviewers' readings. Because the three reviewers agree on every page cited, this is acceptable. However, it left open question 3 unresolved when one look at the page would have settled it (see Section 3).
• Page-offset rule is wrong (minor). mathematician.md says "PDF page = book page + 25 for the body". The offset is actually +23 for book pp.3-16, +25 for pp.17-62, and +27 to +33 from p.63 on, because unnumbered plates sit between pages. The mathematician's individual PDF citations (p.29, p.37, p.83) are correct, but a reader who applies the rule to the photographic, N1 or notes pages will open the wrong page.

3. Missed material and errors found in the text

1. Arbiter open question 3 can be closed from the text, and the answer is "a real inconsistency in the book". PDF pp.37-38 (book pp.14-15) read: quartz lens with a focal length of 12 cm for yellow light, slit at 26.5 cm, focus at "about 13.9 cms". The text then says "By varying the value of l, other values of l' are obtained, and substituting these values in the lens formula, the number 2.93 is obtained ... being the mean value". A second lens (f = 33 cm) gave 2.942. The OCR agrees with the mathematician's figures (26.5, 13.9, 2.93), so the problem is not a misreading. The one printed pair gives n ≈ 1.72, not 2.93. The 2.93 is a mean over unprinted pairs, so the book does not show how it was derived. The arbiter and the primer (Section 7, "Genuinely open", third bullet) should replace "possibly a misreading" with "the printed example pair does not reproduce the printed mean; the other pairs are not given".
2. Missed claim: velocity of propagation. The Preliminary Notice (PDF p.17, book p.viii) says Blondlot's earlier measurements of the velocity of "X" rays (Comptes Rendus t.135, pp.666, 721, 763) "apply in reality not to X rays, but to N rays", and that the velocity "was the same as that of Hertzian waves, and consequently of light". This is a quantitative claim, and historically it is the experiment that started the N-ray programme. It is absent from facts.json key_claims, claims.md, all three reviews, the arbiter and the primer. The only "velocity" mentions in those files concern his separate 1891 measurement of the speed of electric waves. It should be added at least as a sub-claim of claim 1. The 1902 measurement is not in the folder, so its verdict will rest on [web] or [memory].
3. Misreading of the Rubens letter (mathematician, Correction 4). The mathematician says Rubens "made the same observation" for the phosphorescence effect. In context (PDF p.55, book p.30) the sentence follows the thermopile null test: "The action of N rays on this apparatus was absolutely nil ... M. H. Rubens made the same observation, as he was kind enough to write me, his apparatus being much more sensitive even than mine." The phrase "his apparatus being much more sensitive" points to Rubens confirming the null thermopile result, not the brightening. The book never claims a Rubens confirmation of N rays. This misreading did not reach the arbiter or the primer, but facts.json or the arbiter should state the correct reading, because it bears on claim 2's replication column.
4. Corroborators named in the book are under-used. The book cites observations by other people, nearly all at Nancy: - Charpentier: physiological N-ray research, C.R. t.137 p.1049, 24 Dec 1903 (PDF p.73); wet cardboard transmits rays from a knife-blade (note 12). - Bichat: stretched ethyl ether emits N1 rays (PDF p.104); lead's opacity is due to oxide (note 5). - Gutton: action of a magnetic field on a feebly luminous surface (PDF p.106). - Macé de Lépinay: sound vibrations brighten a phosphorescent screen (PDF p.106). - Sagnac: proposed an opposing long-wavelength interpretation from quartz-lens multiple images (PDF p.85).

The primer's evidence table says "No independent replication" for every claim. That is right for independent labs, but the reader should be told that colleagues in Nancy did report confirmations, and they are named in his own book. This is the "over-credulous first" half of the story that the arbiter describes. The contrarian names Charpentier and Sagnac only via [web]. 5. Analogous effects from sound and fields (claim 5 context). Blondlot equates the N-ray effect with the effects of sound waves, a magnetic field and an electromotive force on a dim surface (PDF p.106). He also reports that "if one speaks aloud or whistles, or ... the clenched fist" is brought near the screen, the spots become distinct (PDF pp.113-114). The physicist cites this. The primer mentions only "even the human body", which the book does not say directly; the fist and voice passage is the textual basis. The primer should cite it. 6. N1 rays. These are a second, "inverse" species with indices 1.003-1.041 and wavelengths of about 0.003-0.0083 µm (PDF pp.97-104). Only the mathematician lists them. The primer could note in one sentence that the N1 points continue the same rising n-versus-λ curve (Fig. 10), which makes the anomalous dispersion worse. 7. Small items, for completeness. The thermopile test with the sun is dated 3 July 1904, inside a note dated 20 July 1903 (PDF p.55), so the translation includes later additions. The digest's "temperature rise less than 0.025°C" is a digest error; the text gives 1/4 × 1/15 = 1/60 of a degree, which the mathematician uses correctly.

4. Accuracy spot checks (against _text/N-rays-00bf93.txt)

example.py reproduces the primer's Section 5 numbers exactly. I could not check the CXRO f1/f2 rows against the live Henke table without a web call. They are internally consistent with the Al L-edge jump in f2 at 72.6-72.8 eV.

5. Process

• Source tags. Present throughout and honest. [read] cites book pages; the arbiter and primer mark "[read ..., per reviewers]" where they did not re-open the page. [web] carries URLs. [memory] items carry a confidence. The [calc] tag is used consistently.
• Web checks. Verdict-bearing medium-confidence facts were checked:
• The Al attenuation length and index were retagged [web] by the arbiter from the CXRO al.nff table.
• The Leconte prize was verified by the arbiter.
• Barkla was verified.

The quartz far-infrared index (physicist) and EUV glass reflectance remained [memory, medium] after a failed check, and both are flagged as such. Neither carries a verdict. The Wood test history rests on one secondary source (Wikipedia), and every file says so. - Verdict scale. Applied consistently. The arbiter's rule section, and its downgrade of the mathematician's "internally inconsistent" for claim 3 (after showing that aluminium n can exceed 1 just below the L edge), are good practice. - Arbiter/primer mismatch (minor). The arbiter gives the largest Al index in the band as "about 1.02" (1.023 at 72.0 eV). The primer and example.py give 1.037 at 72.6 eV, right at the edge. The primer value is the later and more careful one, so the arbiter should be aligned. - facts.json not updated with reviewer corrections. Three corrections that every reviewer and the arbiter adopted are missing from facts.json: - "X rays from a tube are essentially unpolarized" is wrong (Barkla). - The quartz index of about 2.9 belongs to N rays, not X rays. - The wavelength unit is derivable, not merely presumed.

facts.json's mainstream_conflict sentence about "indices below 1 for aluminium" is also garbled. claims.md lacks claim 7 (photographic registration), which the arbiter added. - Primer against Step 4 and the reader profile. All nine sections are present in order, with answers at the end. Symbols are defined at first use (λ, E, d, θ_m, r_e, N_a, f₁, f₂, δ, β, L, k, T). Acronyms are defined (EUV, CXRO). The worked example uses real numbers and builds only on Griffiths-level physics. Textbook chapter numbers are given only where they are standard (Griffiths Ch. 9, Schroeder Ch. 7). The Source gap note is repeated, but it repeats the stale "no text layer" wording. - example.py. Runs cleanly with .venv/bin/python and its output matches the primer.

6. Remediation actions

Deep reads (Sonnet; the text is clean OCR prose, not dense mathematics): 1. N-rays-00bf93, PDF pp.16-29: Preliminary Notice and the first paper. Extract the velocity-of-propagation claim (what was measured, how, which C.R. notes), the exact rotation and polarization statements, and what "the experiments are correct" covers. 2. N-rays-00bf93, PDF pp.36-40: quartz-lens index determinations. Record every printed (l, l') pair and the second lens (f = 33 cm, n = 2.942), and state whether any printed pair reproduces 2.93. 3. N-rays-00bf93, PDF pp.54-58 and 102-111: list every other named observer (Rubens, Charpentier, Bichat, Gutton, Macé de Lépinay, Sagnac, Becquerel). For each, say what they reported, and what the Rubens letter actually confirms.

facts.json - Set text_available to "yes (OCR, Step 0)" and update how_read. - Correct the page ranges: body PDF pp.14-116, catalogue pp.118-161. - Add the velocity claim, the N1 rays and the named corroborators. - Fix the X-ray polarization statement, the quartz-index attribution, the wavelength-unit wording and the garbled "indices below 1" sentence.

claims.md - Rewrite the Source gap note: the text is now fully available via OCR; what remains missing is outside sources (C.R. originals, Wood's Nature note, the Berlin reports). - Add claim 7 (photographic registration). - Restate claim 1 as reassigned by Blondlot to N rays, with the velocity measurement as a sub-claim.

arbiter.md (add a "QE revision" section) - Close open question 3: the book's own example pair gives n ≈ 1.72, not 2.93, and no other pairs are printed. - Add a verdict for the velocity sub-claim (likely Tested, not supported, or Never properly tested, depending on the [web] check). - Correct the largest Al index to about 1.04 (CXRO, 72.6 eV). - Note the Nancy corroborators and the correct reading of the Rubens letter.

No existing verdict changes.

primer.md - Update the Source gap paragraph. - Section 7: replace "possibly a misreading" with the resolved lens discrepancy. - Section 6: add a line noting the Nancy colleagues' confirmations (Bichat, Charpentier, Gutton) next to "No independent replication". - Section 1: replace "even the human body" with what the book says (voice, whistling, a clenched fist, PDF pp.113-114), or tag the human-body claim [web] to Charpentier. - Optionally add one sentence on N1 rays continuing the rising n-versus-λ curve.

mathematician.md (optional; it is a Step 2 file) - Note that the "+25" page offset holds only for book pp.17-62. - Correct the Rubens-letter reading.

Recheck

Checked against the current facts.json, claims.md, arbiter.md, primer.md, example.py and deep/N-rays-00bf93.md. Tags: [read] = the files named.

Defects (Sections 1-5)

Remediation actions (Section 6)

Remaining: none that affect facts, verdicts or the primer. One optional, cosmetic item is left in mathematician.md.

Final: PASS