Charles F. Brush: quality engineering (QE, Step 4b)

Date: 2026-10-03. Inputs checked: _text/manifest.json, all 44 digests in digest/, facts.json, claims.md, physicist.md, mathematician.md, contrarian.md, arbiter.md, primer.md, example.py, and spot checks of the extracted text in _text/. No deep/ folder exists: this project was reviewed before digests and deep reads were part of the process, and Step 1 read the key papers directly by OCR of page images.

Verdict: MINOR. The central papers were read in full and the verdicts are well argued. One sub-verdict (claim 4b, "Brush's own pendulum results: Never properly tested") has to be revised because the reviews missed earlier and contemporary pendulum tests of the equivalence principle. That change can be handled in a "QE revision" section of the arbiter, and it does not overturn the headline verdicts. Several smaller omissions and a misleading source-gap note also need fixing.

---

1. Coverage

The manifest lists 12 documents. Eleven were digested. One was skipped:

Browser junk: none. facts.json has 15 entries because it correctly splits the rexresearch bundle into 4 items. Every manifest document has an entry.

Defect, how_read fields are out of date. Every OCR item in facts.json says "how_read": "page images", yet digests now exist for all of them, and Step 1 actually read the OCR text, not the images (for example, the facts entry for the 1922 paper says "rendered ... and ran OCR"). The fields should read "page images (OCR) + digest". For the two bound volumes they should name the exact PDF pages read.

Defect, the source-gap note is misleading. claims.md, physicist.md and arbiter.md all say "the 1922 Proc. APS paper (pp.167-183 only, from a 349-page volume)" and "the 1917 Physical Review paper (pp.228-236 only, from a 615-page volume)". Those page ranges are the complete Brush papers. The rest of each volume is other authors' work. As written, the note implies the key papers were only sampled. They were not. The note should say "complete paper read; the rest of the volume is unrelated, except the Gregorio exchange (Minutes, PDF pp.328-335)". The real gaps (the 1915, 1921, 1923, 1924, 1926-28 papers, the Bureau report and the 1914 manuscript) are listed correctly.

2. Shortcuts

• The central experimental claims (4, 5, 6) rest on full OCR readings of the 1922 pendulum paper, the 1917 steel paper, the 1928 address and the 1929 paper. They do not rest on digests alone, so the shortcuts here are acceptable.
• Claim 3 (the x-term and Laplace) rests on pp.1-16 of the 1914 Discussion. The unread manuscript (pp.17-29) is the only place a formula for x could be. The arbiter turned this into a conditional calculation and said so, which was handled honestly. A deep read of the page images should still be attempted.
• Claim 7 (Eotvos) rests on letters whose last pages are unreadable. The verdict "Never properly tested (never performed)" for Brush's proposed test is supported only by the absence of evidence, and the files should say that plainly.
• Step 1 used no digests (none existed then), but its direct OCR readings of the key papers do the same job as deep reads. I accept them as equivalent.

3. Missed material

1. Earlier and contemporary pendulum tests of the equivalence principle (affects claim 4b). The Wikipedia table of weak-equivalence-principle tests lists Southerns 1910, pendulum, 5×10⁻⁶ and Potter 1923, pendulum, 3×10⁻⁶ [web https://en.wikipedia.org/wiki/Equivalence_principle]. Brush's pendulum effect is about 3×10⁻⁵ in period, or about 6×10⁻⁵ in g. Both pendulum experiments were therefore 10 to 20 times more sensitive than the effect he reported, and they used the same kind of apparatus. Potter's paper came one year after Brush's 1922 paper. My recollection is that it was "Some experiments on the proportionality of mass and weight", Proc. Roy. Soc. A 104 (1923), and that it was a response to Brush that included bismuth [memory, medium confidence; I could not open the Royal Society page (HTTP 403) and my web-search budget is used up, so this is unverified]. The arbiter's sub-verdict 4(b), "Never properly tested (single lab ... no replication)", and the contrarian's "no independent replication on record" both miss these tests. If Potter did test bismuth against other metals in a pendulum, 4(b) becomes Tested, not supported. Even without bismuth, Southerns 1910 shows that a pendulum experiment had already put a limit below Brush's effect before he published. The primer's evidence table needs the same change.
2. The Gregorio priority dispute (Proc. APS 61, Minutes, Stated Meeting of 3 November 1922, PDF pp.328-335). The Marquis Antonio de Gregorio wrote to claim priority for an aether-absorption theory of gravitation dating from 1892 (published in the Atti of the Palermo Academy, 1895). Brush replied with a detailed comparison: Gregorio's aether "tension" is absorbed into matter, while Brush's is isotropic kinetic wave energy [read: _text/Proceedings...txt lines 14090-14289]. No review file mentions this. It is historical context for claim 2: Brush's push theory belonged to a family of Le Sage-type theories, and Brush himself defended what was new in it.
3. A "persistent collateral phenomenon" in the 1922 falling-body apparatus. Brush writes that the drop experiment "developed" a phenomenon "of great interest", which he was investigating and held back for a later paper [read: Proc. APS p.204 (journal p.183)]. It is not mentioned anywhere. This is probably the start of the heat-generation work (claim 5), which links claims 4 and 5 in time, and the primer could mention it.
4. The 1922 paper gives bismuth faster than zinc in the drop test, with a time difference "of the order of a hundred thousandth part of a second" [read: p.204]. In the 1928-29 papers the comparison is different (silicate slower than lead). facts.json mentions the drop apparatus, but no review sets the 1922 bismuth/zinc drop result next to the later silicate results.
5. Brush's frequency reversal. The arbiter already notes the switch from very low frequency (1910-11) to above cosmic rays (1929) as an internal inconsistency. The 1928 address (digest p.5) gives the in-between reasoning, "a wide frequency range like X-rays", which shows the change was gradual. This is a small point the primer could add.
6. The Brush Foundation and T. W. Todd (the brush-paper-spcbru00769 letter). facts.json names Todd. The Foundation's link to Brush's late research funding is not discussed. This is peripheral and needs no action beyond an optional line.
7. The 1921 NYT meeting report also covers Snyder's "variable Planck's constant" and other papers presented at the same session. facts.json notes this. It is context for how the APS meetings of that era received unconventional papers, and is optional for the primer.

4. Accuracy spot checks (against _text/)

The physics was also checked. The arbiter's buoyancy-plus-added-mass formula, (1+k)/2 · ρ_air · (1/ρ_Zn − 1/ρ_Bi), is right for bobs balanced to equal apparent weight in air. Here ρ_air, ρ_Zn and ρ_Bi are the densities of air, zinc and bismuth, and k is the added-mass coefficient (the share of the displaced air that moves with the bob). The restoring forces are then equal, and the inertia differs by the true-mass difference ρ_air·ΔV plus the added-mass term k·ρ_air·ΔV, where ΔV is the difference in bob volume. The text confirms the bobs were matched by weight ("made approximately the same in weight, about 1.377 kg", line 7933). The text does not say whether the weighing was done in air. In 1922 it almost certainly was [memory, high confidence], but a deep read should confirm what the paper says.

5. Process

• Source tags: Present and generally honest throughout. [memory] claims carry confidence levels, Step 1 marks unread pages "[not read]", and the arbiter labels its own arithmetic [calc]. A good practice: the arbiter states when a web check failed (Bureau report, cylinder added-mass coefficient). One issue: the "page images" label in facts.json overstates what was done (OCR text was read, not the images), as noted in section 1.
• Web checks: The physicist, contrarian and arbiter each made several [web] checks (Eotvos, MICROSCOPE, speed of gravity, added mass, martensite). The arbiter has 10 [web] tags, within its budget of 8 to 10. Gap: no reviewer looked for historical tests of Brush's specific claim from the 1910s-1920s, even though the verdict on claim 4(b) turns on whether any exist. That is the main process miss.
• Verdict scale: Applied consistently, with split verdicts used properly. The arbiter's remark that the reviewers' labels for claim 7 are both partly right is well reasoned. Claim 6 ("Plausible, untested"): the physicist and contrarian say an independent confirmation probably exists in metallurgy but could not source one. The verdict is correct as written, but it should be rechecked on the web: the heat released during room-temperature ageing of martensite is standard metallurgy [memory, medium confidence].
• Primer: Follows STEP 4 order (sections 1-9 plus answers). Symbols and the added-mass idea are explained, the worked example uses real numbers from Brush's tables, and the evidence table is present. It suits the reader profile: no tensors or GR formalism, and general relativity is mentioned only for context.
• example.py: Runs cleanly with the venv Python. It reproduces 1 in 34,865 (Brush: 35,000), 40.39e-6 (Brush: 40.3e-6), the geometric corrections 9.01e-6 / 2.53e-5 (Brush: 9.0e-6 / 2.5e-5), bob masses from their dimensions within 1%, and the air prediction of 3.4-4.5e-5. The output matches the primer.

6. Remediation actions

Deep reads (deep/{slug}.md, tagged [read: p.N]):

1. Proceedings-of-the-American-Philosophica-771a0a, PDF pp.184-204, Opus. How were the bobs matched: weighed in air on a balance, or by some other method? What exactly does Brush say about vacuum, buoyancy and air (p.201: "To this I can confidently answer no")? What is the "persistent collateral phenomenon" of the drop apparatus (p.204)? List any internal contradictions (for example, the reversed orderings in the short-pendulum runs). This confirms the premise of the arbiter's added-mass verdict.
2. discussion-kinetic-theory-gravity-spcbru01027-c65704, pp.17-29, Sonnet, from page images (the OCR is unreadable). Are these drafts of the printed text, or do they contain new material, especially a formula or numerical value for x in 1/D^(2−x), or an orbit calculation? If a formula exists, the mathematician's or arbiter's perihelion calculation for claim 3b should be redone with it.
3. Proceedings-of-the-American-Philosophica-771a0a, PDF pp.328-335, Sonnet. Summarise the Gregorio priority letter and Brush's reply: what each theory assumes, the differences Brush claims, and the dates.
4. Evidence-for-Kinetic-Theory-of-Gravitation-1-9d8a5c, pp.4-7, Sonnet. Close the passage Step 1 says it did not read: the calorimeter descriptions, the Bureau of Standards statement, the silicates tested and any numbers.
5. brush-miller-letters-spcbru00840-0c03d5, pp.4-7, Sonnet, from page images. What do Miller's later letter(s) say? Did he agree to attempt the Eotvos-type test, or decline?
6. Spontaneous-gen-of-heat-in-hardened-steel-Physical-review-2437d4, PDF pp.94-97, Sonnet. The APS meeting abstract of hardened-steel paper II: does it add numbers or claims beyond the full paper?

Web checks (arbiter, during the QE revision): confirm Potter 1923 (Proc. Roy. Soc. A 104): which materials were tested (bismuth?), the precision, and whether it cites Brush. Confirm Southerns 1910 (Proc. Roy. Soc. A 84): which materials and what precision. Look for any 1920s published response to Brush's falling-body or heat claims, and for metallurgical confirmation of room-temperature heat release from fresh martensite.

Changes to later files:

• facts.json: Update how_read for all OCR items to "page images (OCR) + digest". Name the PDF pages read for the two volumes. Correct the 1922 paper's date (written April 1922, printed 16 Nov 1922). Add the Gregorio exchange to the Proceedings entry. Add the "collateral phenomenon" and the 1922 bismuth-faster drop result. Record that the truncated 1928 passage is now covered by the digest and the deep read.
• claims.md: Rewrite the source-gap note: the 1922 and 1917 papers were read complete, and the remaining volume pages are unrelated. Keep the list of missing papers. Add the 1914 manuscript status after deep read 2.
• arbiter.md: Add a "QE revision" section. (a) Claim 4(b): add Southerns 1910 and Potter 1923 [web]. If Potter tested bismuth, change 4(b) to Tested, not supported; otherwise change it to "Tested, not supported in general pendulum form (10-20 times more sensitive than Brush's effect); Brush's specific pairing never replicated". Update the contrarian-category column ("not dismissed; contradicted by contemporaneous pendulum tests"). (b) Claim 3(b): revise only if deep read 2 finds a formula. (c) Claim 7: add what Miller's later letter says. (d) Add a line on the Gregorio dispute under claim 2.
• primer.md: Update the evidence-table row for claim 4 and section 7 (takeaways) to mention Potter and Southerns. This is instructive: in the 1920s a better pendulum was built, and it did not see the effect. Add one sentence of history on Gregorio and Le Sage-type priority, and one on the 1922 drop experiment and the "collateral phenomenon" that led to the heat work. No change to the worked example or to example.py.
• Recheck (Sonnet): Append a "Recheck" section to this file stating which of the defects above were fixed.

---

Recheck

Date: 2026-10-03. Checked against the current facts.json (valid JSON), claims.md, arbiter.md, primer.md, example.py (runs cleanly) and deep/ (5 files).

Defects

Remediation actions

Open issues: (1) Gregorio deep read not done. (2) Web verification of Potter/Southerns details, 1920s responses and martensite heat release not done. Neither changes a headline verdict. Both are stated openly in the files.

Final: OPEN (2 issues)

---

Recheck 2

Date: 2026-10-03. Re-checked against the current facts.json (valid JSON), claims.md, arbiter.md, primer.md and deep/ (5 files).

Open issues: (1) Web verification of Potter/Southerns details, 1920s responses and martensite heat release not done.

Final: OPEN (1 issues)