QE (Step 4b): Poincare-Henri, "The New Mechanics"

QE agent: Opus. Date: 2026-10-06. Inputs read in full: _text/manifest.json, _textprep.log, digest/TheNewMechanics-7c45ca.01.md (the only digest), deep/TheNewMechanics-7c45ca.md, facts.json, claims.md, sites.md, all five research/ notes (research/files/ is empty), physicist.md, mathematician.md, contrarian.md, arbiter.md, primer.md and example.py. I also read the extracted text _text/TheNewMechanics-7c45ca.txt in full, and I rendered PDF p.6 to an image (/tmp/qe-poincare/page-06.png) to check the one figure the text layer garbles. Tags: [read: p.N] = PDF page N of the project document (journal page 384+N); [calc] = my own arithmetic (.venv Python); [memory] with confidence.

Verdict: MINOR

The coverage is complete and the document was read in full. The central verdicts hold up against the text, and I found no error that would change a central claim or verdict. The defects can all be fixed by updating facts.json, the arbiter (a "QE revision" section) and the primer: one stale fact in facts.json, three places where the verdict scale is applied loosely, a "Refuted by calculation" verdict that shows no calculation, a few Investigator leads and source passages that dropped out of the arbiter and primer, and some undefined acronyms in the primer. No deep reads are needed. The project has one 11-page document, it was deep-read in full, and I re-checked the passages myself. The open records (Sechs Vortrage 1910, the HSNS 2023 study, Walter, Wacker) are web sources, not manifest documents, so they are a job for the next Investigator run (listed below).

1. Coverage

Nothing was skipped. _textprep.log reports 1 document and 1 chunk, which matches the manifest. No defect.

Digest quality note (expected at this stage, already caught downstream): from the contraction passage onward the digest's page labels run one page late. It puts the contraction at p.7 (the text has it at p.6), the atom at p.8 (text p.7), Mercury at p.10-11 (text p.10) and Leverrier at p.11 (text p.10). The deep read, facts.json, claims.md and the reviews all use the correct pages, so nothing downstream inherits the error.

2. Shortcuts

None. The single key document was read in full (text layer plus page images where the layout garbled the text). No central claim rests on a digest alone. The essay has no equations, so no page-image reading of formulas was needed beyond the p.2 figure and the p.6 number. No defect.

The real limit of the project's evidence base sits elsewhere, and the arbiter and primer state it correctly. All the context comes from web texts the Investigator read, not from files in the project: the date, the Lille twin lecture, the 1906 memoir, de Sitter, Bucherer and Lorentz 1900. research/files/ is empty. That is acceptable for this project, but those texts should be saved when the next Investigator run reaches them (see section 7).

3. Accuracy spot checks (against _text/TheNewMechanics-7c45ca.txt and the page image)

example.py runs cleanly (.venv/bin/python example.py, exit 0). Every number it prints matches the primer's text and tables: A1 4.934e-9 = 1/2.027e8, 6.29 cm, 7.62e-7 at 370 km/s; A3 0.43 fringe; A4 0.145 fringe; B1 to B6. I also checked the primer's five "Check yourself" answers (0.6c gives 0.832c; 5.0e-7 is about 101 times the Earth figure; 7.15/43 = 0.166 with a margin of about 36"): all correct [calc].

4. Missed material (in the source or the Investigator's notes, missing from downstream files)

A. In the essay but absent from every review, the arbiter and the primer: 1. Leverrier's undiscovered planet and "only a bird" [read: p.10]. Before the relativistic correction, Poincare tells the story of the ad hoc fix for Mercury: Leverrier attributed the anomaly to an undiscovered planet, an amateur astronomer thought he saw it cross the Sun, and "it is unhappily certain that this planet perceived was only a bird." It is in the digest, but no file after facts.json carries it. It is pertinent and instructive: it is the 19th-century ad hoc hypothesis that sits beside the contraction hypothesis, which the primer itself calls ad hoc (section 7, "Treating one null result as verification"). The planet's later name, Vulcan, and the amateur, Lescarbault in 1859, are [memory, high] and not in the text. 2. The 1909 atom model [read: p.7]: "a positive central mass around which circulate a great number of negatively charged electrons", with the atom as a "universe" of planets around "tiny suns". This is two years before Rutherford's nuclear atom of 1911 [memory, high]. Only facts.json mentions it. It is a historical point worth one sentence in the primer: the essay describes a nucleus-like model before the experiment that established the nucleus. 3. X-rays from cathode-ray impact [read: p.9]: cathode rays strike the anticathode, giving an "electromagnetic agitation which many physicists identify with the Rontgen rays". Lorentz's electron-gas account of incandescence ("a swarm of gnats in a jar") is on p.8-9. Both are early forms of ideas that became standard: X-rays produced by decelerating electrons (bremsstrahlung), and the free-electron (Drude-Lorentz) picture of metals [memory, high]. None of the reviews judges them, and the primer's "What is solid" list could include them. The primer mentions X-ray tubes only generically (section 2).

B. Investigator findings that reached the reviews but not the arbiter or primer (see Retention, section 5): the Lille separation of inertial mass from gravitational mass; the Lille aberration test; Weber's law.

C. facts.json is stale on the date. The "year" field says "unknown from the document" and the sources line still carries "[memory] medium ... printed in The Monist about 1910". All three reviews and the arbiter correct this to: lecture of 28 April 1909 at Gottingen, French text in Sechs Vortrage (1910) pp. 49-58, English in The Monist 23(3), July 1913, pp. 385-395 [web, en.wikisource and the JSTOR snippet]. facts.json has not been updated.

5. Retention check

The rule: pertinent material from the sources or the Investigator's findings must not be left out of the arbiter or the primer because a reviewer judged it negatively. Overall this project does well. The material most damaging to Poincare's own presentation is kept in both the arbiter and the primer, with sources: Kaufmann's own anti-Lorentz conclusion, Bestelmeyer's doubts about Bucherer's errors, Zahn-Spees 1938, and Bucherer's list of Kaufmann's errors, described as a competitor's characterization. So is material that favours him: the reply to Laplace, Mercury's correct direction confirmed by three independent calculations, the Lorentz ether as empirically equivalent to special relativity rather than refuted, and the non-electromagnetic stress in his 1906 memoir. The Truman State offprint, the geocentrist book (reported as that book's characterization, beside Poincare's own disavowal at Lille), the missing "6"" and the unnamed Einstein are all analysed as absence of evidence. No item was dropped because a reviewer judged it negatively. Items dropped by omission (minor defects):

6. Process

• Source tags. They are present and honest throughout, with [read: p.N] page numbers checked against the text. The arbiter, contrarian and primer add a [calc] tag. It is not in the prompt's list, but it is used consistently and honestly for arithmetic, so it is acceptable. Medium-confidence [memory] items are flagged as not verdict-bearing (the Michelson-Morley 0.02 fringe, the Paris-Berlin distance, Poincare's "gravific waves"). OK.
• Web checks. Physicist: 3 checks plus the Investigator's notes. Contrarian: 3. Arbiter: 5 of 8 (Wikisource text, Kaufmann-Bucherer-Neumann, electromagnetic mass, Kennedy-Thorndike, plus one null attempt). The mathematician ran none, and states that no verdict turns on a medium [memory]. One exception: its claim 4 argument rests on "Kaufmann's early interpretation" at [memory, medium-high]. The arbiter's web check 2 covers it. Minor; no action beyond what the arbiter already did.
• One verdict-bearing claim without a check or calculation. The physicist dismisses Lorentz's "Franklin" charge-asymmetry gravity as "refuted by calculation" on [memory, medium], not web-checked, with no calculation. The arbiter did not adopt this as a verdict, so no verdict is affected. If the point is kept, it needs a source or a calculation.
• Verdict scale consistency (the main process defects): 1. Claim 1, arbiter and primer evidence table: "the essay's statement that the experiments had 'ascertained' it in 1909 ... Historically superseded but instructive as history of evidence." The scale classifies claims about nature. An overstatement about the 1909 state of the evidence is not a claim that was "a reasonable position once; replaced since". Fix: drop the scale label and state it as a documented overstatement: the 1909 data were contested (Kaufmann, Bestelmeyer, Zahn-Spees 1938). 2. Claim 3, "Michelson's experiments verify the deformation": Tested, not supported. The scale defines this as "controlled experiments were run and did not support it". The Michelson-Morley result is consistent with the contraction; it does not fail to support it. It is just not an independent test of it. The arbiter itself says "overstated rather than disproved", which contradicts the label. Fix: call it an overstatement of what a null result can show (no scale verdict), or "Not testable as stated" for "a single null result isolates the deformation". Keep the substance: the null result is compatible but does not isolate the contraction, and Kennedy-Thorndike needs time dilation as well. Update the primer's evidence table to match. 3. Claim 5b(i), "Refuted by calculation", but no calculation shown. The scale says "Show the calculation." Neither the physicist nor the arbiter computes anything for the electromagnetic case. One calculation that supports the verdict, which the arbiter should add [calc, mine]: give the Earth a generous net charge equal to the fair-weather field's surface charge, q = eps0 x 100 V/m x 4 pi R_E^2 = 4.5 x 10^5 C (the 100 V/m figure is [memory, medium]). Its orbital acceleration is a = v^2/r = 5.93 x 10^-3 m/s^2. The Larmor power is P = q^2 a^2 / (6 pi eps0 c^3) = 1.6 x 10^-9 W. The orbital binding energy G M m / 2r is 2.65 x 10^33 J, so the decay time is about 5 x 10^34 years: about 10^17 times longer than the essay's 10^17 years, and far longer than gravitational-wave decay (10^23 yr). This keeps the arbiter's distinction intact: within Lorentz's residual-force model the claim remains "Not testable as stated". 4. Claim 4: the arbiter combines "Tested, not supported (indirectly)" with "Historically superseded". The scale allows a split verdict, and the arbiter gives its reasoning, so this is acceptable. No action.
• Primer against STEP 4 and the reader profile. All nine sections are present, in order. The worked example is real and reproduced by example.py, the answers are at the end, and the textbook chapters are named only where certain (Griffiths ch. 12, Feynman I.15-16 and II.28 are correct [memory, high]). The defects are acronyms and terms used without definition, against the reader profile:
• "GR" is used in sections 5 and 6 and is never introduced as "general relativity (GR)" in the primer.
• "LIGO" (section 3) is undefined.
• "Higgs coupling" (section 4 table) and "quarks" (section 3) assume quantum field theory, which the profile does not assume. They need one plain-language clause each or should be removed.
• "AFAS" and "MNRAS" (section 9) are undefined.
• "SI" (section 2) is undefined. This one is trivial.
• Section 1 says the new mechanics is one "which he credits to Lorentz and Max Abraham". The essay names them as the "destroyers" of Newtonian mechanics [read: p.1]. Their electron models were rivals, and Abraham's lost to Lorentz's in Bucherer's 1908 data. Reword so the reader does not take Abraham as a co-author of the Lorentz theory.
• example.py runs: yes (exit 0), and its output matches the primer.

7. Remediation actions

Deep reads: none. The only manifest document was deep-read in full (pp.1-11 with page images of pp.2 and 6), and QE re-read every passage at issue directly from the extracted text and a page image.

facts.json 1. Set "year" to "Lecture 28 April 1909 (Gottingen, Wolfskehl lectures); French in Sechs Vortrage (1910) pp. 49-58; English tr. Halsted, The Monist 23(3), July 1913, pp. 385-395 (posthumous)". Replace the "[memory] medium ... about 1910" source line with the [web] sources (en.wikisource The_New_Mechanics_(Göttingen); the Lebon bibliography and JSTOR snippet in research/2026-10-06-google.md). 2. Add to the summary: the Leverrier planet and "only a bird" passage [read: p.10]; the Rontgen-ray and incandescence passages [read: p.8-9] (the atom model is already there).

claims.md: no change of claims. Optionally add one line noting the date correction.

arbiter.md: add a "QE revision" section 1. Claim 1: replace the "Historically superseded but instructive as history of evidence" label with a plain statement that the essay overstated the 1909 evidence. The verdict stays Established. 2. Claim 3: replace "Tested, not supported" for "Michelson verifies the deformation" with an overstatement note (or "Not testable as stated" for the isolation claim), keeping the Kennedy-Thorndike reasoning. The other parts of the split are unchanged. 3. Claim 5b(i): add the Larmor calculation in section 6 above, so "Refuted by calculation" shows its calculation. The verdict is unchanged. 4. Claim 5: add the Lille inertial-versus-gravitational-mass passage ("aucun moyen de décider") as a question Poincare recorded as open, and Lorentz's remark on Weber's law as a related lead. 5. Claim 2: add the Lille aberration-test passage (Poincare certain in advance of compensation) beside the "absolute value" versus "postulate open to test" point. 6. Optional: one line on Voigt 1887 under absence-of-evidence item C.

primer.md 1. Section 1: reword "credits to Lorentz and Max Abraham" (the essay names them as attackers of Newtonian mechanics). 2. Define GR, LIGO, AFAS and MNRAS on first use. Replace or explain "Higgs coupling" and "quarks" in plain language. 3. Section 4 or 7: add the Leverrier-planet/"bird" story as the classic ad hoc hypothesis that failed, set beside the contraction hypothesis. 4. Section 7, "What is solid": add the essay's correct early pictures of X-rays from cathode-ray impact and of electrons in a hot body (one line each, [read: p.8-9]). Add the pre-Rutherford nucleus-like atom as a historical note [read: p.7]. 5. Section 7, "What is genuinely open": add Poincare's own open question at Lille, whether the attraction coefficient depends on velocity (inertial versus gravitational mass), and note that it later became the equivalence principle. Mention Weber's law as related work. 6. Section 6 evidence table: carry over the arbiter's revised labels for claims 1 and 3 and the calculation reference for 5b(i).

Next Investigator run (not part of this remediation; recorded so it is not lost): Sechs Vortrage (1910) pp. 49-58, which decides whether the Gottingen French had the "6"" and Bucherer; the JSTOR Monist 23(3) issue; the HSNS 2023 "Hazy Spots on Photographic Plates" study; Lorentz, Phys. Z. 11 (1910); the Wacker dissertation; Walter on Poincare's two gravity laws; Weber's law and Mercury; the Truman offprint's volume line. Save the Lille, Palermo-memoir, de Sitter and Bucherer texts to research/files/ where they can be downloaded, so the project's context no longer rests on web pages alone.

Recheck (Step 4b, 2026-10-06)

I re-read facts.json, arbiter.md (including its "QE revision" section) and primer.md, and checked each defect from sections 4 to 7 against the current text.

Housekeeping note, not a defect: the arbiter's "Housekeeping and still open" section still says facts.json says "about 1910". That line is now out of date, because facts.json has been fixed. It affects no verdict.

The next-Investigator-run records (Sechs Vortrage, the HSNS study, Wacker, Walter, saving texts to research/files/) were never part of this remediation and remain open as leads.

Final: PASS