QE (Step 4b): Kennard, "On Unipolar Induction" (1917)

QE agent: Opus, 2026-10-04. Inputs checked: _text/manifest.json, digest/ (1 file, read in full), deep/ (1 file), facts.json, claims.md, sites.md, research/ (notes and the saved 1912 and 1917 IA OCR texts), physicist.md, mathematician.md, contrarian.md, arbiter.md, primer.md, example.py. I also re-read the extracted text of pp.9-12 and rendered page images of pp.5-8 at 150 dpi in /tmp/qe_kennard/ [read: page image pp.5-8].

Verdict: MINOR

The work is complete and of high quality. The one source document was fully read, the deep read caught every load-bearing OCR error, and the reviews and arbiter are careful and consistent. The primer follows Step 4. example.py runs cleanly. The defects are fixable by updating facts.json, claims.md, arbiter.md and primer.md. One open item could change a verdict. It concerns the theoretical charge -6.22 V0, on which the claim-2 "agreement to a few percent" rests. The primer found that Kennard's printed solution is not reproduced by a simple model, but the arbiter's verdict does not take that into account. If a re-solution shows the true charge is much smaller than -6.22 V0, then the claim-2 verdict changes, and only that one. Claims 1, 3, 5 and 6 rest on "charge versus no charge" and would stand. I therefore grade this MINOR, with a stated escalation condition (see R1).

1. Coverage

• Every manifest document was digested. Nothing was skipped. The last half page (p.12) is the start of an unrelated Thornton article, which was correctly ignored.
• Stale Step 1 files (defect, minor). claims.md still opens with "pages 10-12 ... rest on the digest alone, so the aether claims below should be confirmed against the text". facts.json how_read says the same. Both were closed by the reviewers and the arbiter (arbiter "Source gap note"). Neither file was updated, and neither carries the arbiter's corrected claim list (arbiter §3).

2. Shortcuts

• The central experimental claims (1-4) rest on a page-image deep read of pp.1-9. Acceptable.
• The theoretical claims (5-6, pp.10-12) did not get a formal deep read. Instead the reviewers read the IA OCR, which is clean enough to quote. The §7 equations (1) and (2) on journal p.188 (PDF p.10) are garbled in both OCRs ("div (Kf)=dop", "E "). No one reports checking them against the page image of p.10. The reviewers reconstruct Barnett's assumptions as f = e + E + [v x B]/c and div(Kf) = 4πρ. That reconstruction is very likely right, because it matches Kennard's prose. But the claim-5 verdict quotes these equations, so they should be confirmed from the image (deep read D3). Minor.
• The −6.22 V0 charge was never independently reproduced. The reviewers and the arbiter checked every number downstream of it. The primer (Section 5, Step 6) and example.py Steps 5-6 then went further and found a real tension:
• Evaluated numerically, Kennard's printed outer-space solution reproduces his outer-face charge (−7.58 against the printed −7.62 V0), so the formula was read correctly.
• However, it places the inner cylinder at about −0.53 V0. The rigid-rotation rule Kennard states himself (potential = Nn, proportional to r²) puts it at about +0.054 V0. [calc: example.py Step 6 output: V(0.77, 0) = −0.5297 V0; potential drop across the gap 1.525 V0 printed against 0.946 V0 naive]
• A simple infinite-coax estimate gives an outer-face charge of −4.89 V0, which is 64% of Kennard's value.
• The paper does not print its boundary values. The arbiter lists only a generic open question ("Is −6.22 V0 right to 0.5%?"). Its claim-2 verdict, "Established ... magnitude to a few percent", does not mention that the primer's own check fails by about 35%. Nor does it consider what follows if the charge is substantially smaller: the measured ratio 0.242 would then exceed the Lorentz prediction by tens of percent. The primer argues that "the experiment itself sides with Kennard's number", and labels this an inference. That argument is partly circular, because it assumes the theory being tested. This is the most important shortcut in the project (R1, D1).

3. Missed material (from my own reading of the digest, the text and the page images)

1. The sign of the solenoid-alone residual. In the "Both" rows the electrometer sign equals sign(S·I): +14.4, −15.6, +16.4, −18.0 [read: page image p.8]. The "Rev. av." of +1.2 mm for the solenoid-alone rows uses the same weighting. Under the moving-line theory, turning the solenoid in sense S is equivalent to turning the condenser in sense −S. Its prediction for these rows would therefore be about −15 mm, of the opposite sign to the residual found. If S has the same meaning in both blocks of the table (an assumption: Kennard does not say), then the +1.2 mm residual cannot be a partial moving-line effect. It also bears on Su's end-effect proposal, whose sign is not stated. No review makes this point [inference from read data, to be confirmed by D2].
2. The solenoid-alone G column is not "0" row by row. The arbiter (claim 3) and the primer (Step 8, evidence table) say "G = 0 in those rows". The page image shows G = +0.5, −0.3, −0.6, +0.2 mm, with a reversing average of 0 [read: page image p.8]. Their conclusion (no speed record, because the Faraday disk on the condenser shaft is not turning) stands. The wording should be "G ≈ 0 (±0.6 mm; reversing average 0)".
3. Raw triples for the solenoid-alone rows show large drifts. They are +8.5, +4.8, +3.7 and +22.4, +13.8, +1.8 mm [read: page image p.8]. Individual readings in the null rows thus move by up to 20 mm before drift removal. This is the real noise floor behind the +1.2 ± 0.8 mm null, and it belongs in the primer's discussion of claim 3.
4. Hertz's theory. It is the third alternative. Kennard says the condenser should charge "under either Lorentz or the moving-line theory, but not according to Hertz's" [read: deep read §1, p.2]. The primer quotes Kennard 1912 ("not capable of distinguishing" Lorentz from Hertz) but never says what Hertz's theory is or what it predicts. For this reader the name is undefined (also a reader-profile defect).
5. Kennard's answer to Barnett's 1912 objection and his own concession. The answer: induced and permanent magnetism would have to differ radically, and "the remanence in the bar employed amounted to over 25 per cent". The concession: the 1912 inner screen "might conceivably act so as to cut off the effect", which is the very point the 1917 experiment was built to settle [read: pp.1-2]. The arbiter and primer give only Barnett's side ("his technical objection ... was fair").
6. Kennard's characterisation of Fehrle. He says Fehrle's work "is marred by several results which must certainly be wrong: one of his results even contradicts Faraday's law for closed circuits!" [read: p.2]. The primer cites Fehrle only as a confirmer, and does so inside a quote. Retain this as Kennard's characterisation of a third party's work, which no one has checked.
7. The amber-ring and connecting-wire corrections. These are a +0.03 V0 correction (from Kennard's 1912-13 Phys. Zeits. dielectric papers) and a −0.02 V0 correction, which "just about cancel" [read: pp.6-7]. They are part of the −6.22 V0 chain, and the contrarian notes the cancellation is asserted rather than shown. They are absent from the arbiter and the primer. They matter for R1.
8. The 1912 paper already contains the Laplace/divergence argument. The Investigator found this (sites.md, priority 1), and I confirmed it: "E' in general fails to satisfy Laplace's Equation" [read: research/files/Kennard-1912 IA OCR line 136]. The primer says Kennard's aether reading predates the 1917 data, but does not say the §7 argument does too.
9. Kennard's 1912 subtracted effect. In 1912 Kennard subtracted an "effect apparently due to electromagnetic induction when the magnetization was reversed" before quoting 0.0 ± 0.24 [read: 1912 IA OCR lines 195-199]. Su 2005 builds on this point. It is kept in the contrarian review, but the primer's Su entry mentions only the end-effect proposal.
10. Minor numbers. The galvanometer swing time is 15 s on the page image [read: page image p.5]. The deep read and the digest say 14 s, the arbiter caught it, and facts.json and the deep read are uncorrected. Kennard's notation is "Np" with p = rev/s. The deep read writes "Nn", which is harmless.

4. Accuracy spot checks (against page images and extracted text)

5. Retention check

The arbiter and primer keep the major negative or contested material well. Examples: Barnett's "fallacies" remarks as a rival's characterisation, Su 2005 with his local-ether stake noted, Baumgärtel and Maher's closing-wire caveat, the Weber/Ampère school with the "fringe" label attributed, the Chyba Earth-rotation dispute, the Absence-of-evidence table with suppression (e) assessed rather than dismissed, the rejected and abandoned data sets, and the accidental 10 kΩ resistance. The following pertinent items from the Investigator or the reviews were dropped from both the arbiter and the primer:

1. The Semantic Scholar abstract claiming "undisputable experiments that categorically disprove absolutistic Kennard-Bartlett-type interpretations" (research/2026-10-04-crossref-semanticscholar.md; kept in mathematician.md §6). This is evidence against Kennard's interpretation. SOURCE HONESTY says evidence against a claim is also stated. It should be kept as that source's characterisation, with the note that the paper is unidentified.
2. Kholmetskii 2003 (Am. J. Phys. 71, 558) and Chyba's remark that an "attempted theoretical refutation" was "shown to be incorrect due to calculational error" (sites.md priority 3). This is a modern re-analysis of the Barnett experiment, with a disputed exchange around it. It is mentioned only in physicist.md as "not read".
3. Miller 1981 and Pegram's 1917 "reminder" to Barnett and Kennard (sites.md, "not reached"). It bears on the priority question the arbiter does raise (open question 2), but the lead itself is not recorded.
4. The 1924 University of Wisconsin B.S. thesis on unipolar induction that cites Kennard (Google Books KalsAAAAMAAJ, full view, not read). This is historical-record uptake, and it is relevant to the "line goes dark" analysis.
5. Items 3.5 and 3.6 above (Kennard's rebuttal of Barnett's objection; his characterisation of Fehrle): Kennard's side of the dispute is shorter than Barnett's.

None of these omissions changes a verdict. Items 1 and 2 are the only ones that could look like selective retention, and they cut in opposite directions (1 against Kennard; 2 against a refutation of the Barnett-type result). Both should be restored.

6. Process

• Source tags. Present and honest throughout. [calc] is used for recomputation, which is a sensible extension. Interested parties are flagged correctly: Barnett as a rival, Chyba as having a stake in Earth-rotation power, Su as having a local-ether theory. One wrinkle: the arbiter quotes Barnett 1919 by page "[read: p.1510]" and "[read: p.1511]", while the contrarian cites the same report at "p.735-736" and "p.737-745 of the extracted text". These are different pagination schemes (journal pages against extracted-text lines). The primer should give one consistent citation.
• Web checks. The physicist left Barnett's affiliation as an unverified medium-confidence [memory] item, and the arbiter web-checked it (2 of 8 calls). The mathematician and contrarian state that no verdict depends on medium- or low-confidence [memory]. I agree. Acceptable.
• Verdict scale. Applied consistently, with split verdicts used appropriately. Arbiter 6b ("Plausible, untested for any alternative other than the mainstream reading") is an awkward use of the scale: it labels a sub-claim that is really a remark. Acceptable, but the primer could say in words what it means. The physicist's claim-2 phrase "order-of-magnitude (about 20%)" is looser than the other reviewers' "few percent". The arbiter resolved this.
• Primer against STEP 4. All nine sections are present, in order. Symbols and units are defined (Gaussian units, e.s.u., div, Bessel functions). The worked example is real and reproduced in example.py. The reader-profile gap is "Hertz's theory", which is used undefined (item 3.4).
• example.py. Runs cleanly in the venv (Steps 1-10, no errors). Outputs match the primer's numbers (1606 G, 55,602 Mx, 0.0183 V, 14.83 mm, 0.2308, +4.8% = 2.1 SE, 2ωB, ω²r).
• Arbiter internal consistency. The arbiter says the contrarian's "under one standard error" is wrong. Its own recomputation (2.1 SE) supports that, and the primer adopts it. Good.

7. Remediation actions

Deep reads (most important first; slug On-Unipolar-Induction-60158c)

Calculation (not a deep read)

• R1 (escalation condition). Extend example.py with a finite-difference solution of the two cylindrical shell spaces, using the boundary assignment D1 establishes, and compare it with −7.62, +1.40 and −6.22 V0. If the re-solved charge is within about 5% of −6.22 V0, the claim-2 verdict stands. If it differs by more than about 15%, the measured ratio 0.242 no longer agrees with the Lorentz prediction to "a few percent". The arbiter must then revise claim 2: the qualitative charge/no-charge test still holds, but the quantitative agreement becomes "unresolved", or an anomaly to report. The primer must be revised to match. This alone does not affect claims 1 and 3-6.

facts.json

• Update how_read: pp.10-12 were read in the extracted text and the cleaner IA OCR by every reviewer (and in page images once D3 is done).
• Correct the galvanometer swing time to 15 s.
• Add to the summary: the solenoid-alone G values (+0.5, −0.3, −0.6, +0.2), the 1912 paper as the origin of the Laplace argument, and the open status of −6.22 V0.

claims.md

• Remove the "pp.10-12 rest on the digest alone" caveat.
• Adopt the arbiter's corrected claim list (arbiter §3), for example "not divergence-free (div(v x B) = 2ωB)" and "apparently the only remaining alternative".
• Add a line that −6.22 V0 is not independently reproduced (R1).

arbiter.md: add a "QE revision" section

• Claim 2: record the primer and example.py finding (the printed solution implies a 1.53 V0 gap drop against 0.95 V0 for the naive rule, and the infinite-coax estimate is −4.89 against −7.62 V0). State that the "few percent" agreement is conditional on −6.22 V0. Revise the verdict after R1 if it fails.
• Claim 3: correct "G = 0 in those rows" to "G ≈ 0 per row (reversing average 0)". Add the sign observation (item 3.1) once D2 confirms the convention, and the single-reading noise level (item 3.3).
• Claim 5: note that the divergence argument first appears in Kennard 1912 (item 3.8), and confirm the equations from D3.
• Retention: add the Semantic Scholar "categorically disprove" abstract (as that source's characterisation; paper unidentified), Kholmetskii 2003 and Chyba's "calculational error" remark, Miller 1981 on Pegram's reminder, and the 1924 Wisconsin thesis. Add Kennard's rebuttal of Barnett's 1912 objection and his Fehrle criticism (from D4).

primer.md

• Section 2 or 3: define Hertz's theory in one or two sentences and add it as a column (or a note) in the Section 4 comparison table.
• Section 5, Step 6: link the −6.22 V0 tension explicitly to the claim-2 verdict, and state the consequence either way. Remove or qualify the partly circular "the experiment itself sides with Kennard's number". Update after R1.
• Section 5, Step 8 and the evidence table: "G ≈ 0" wording, the noise level of single readings, and the sign of the +1.2 mm residual (after D2).
• Section 6, "Related and later work": add the Semantic Scholar "categorically disprove" characterisation, Kholmetskii 2003 and the Chyba "calculational error" remark, Miller 1981/Pegram's reminder, and the 1912 origin of the divergence argument. In the Barnett-1912 entry, add Kennard's own answer (the remanence) and his concession about the 1912 screen. Add Kennard's characterisation of Fehrle.
• Make the Barnett 1919 page citations consistent.

Recheck

Recheck by QE agent, 2026-10-04, against the current claims.md, facts.json, arbiter.md (new section 9, QE revision), primer.md and example.py (Step 11 added). I did not re-run example.py: the default python3 here lacks sympy, so a run was not possible in this environment. I checked by reading the files and grep.

New defects introduced: none found. The stated verdict changes are consistent across claims.md, arbiter.md and primer.md.

Final: PASS