Builder-G: Quality Engineering (Step 4b)

Project: one document, G. Builder, "Ether and Relativity", Aust. J. Phys. 11, 279-297 (1958; manuscript received 20 Nov 1957), Ether-And-Relativity-1957.pdf, 19 pages.

Files checked: _text/manifest.json, digest/Ether-And-Relativity-1957-51e9ba.01.md and .02.md (both read in full), facts.json, claims.md, physicist.md, mathematician.md, contrarian.md, arbiter.md, primer.md, example.py. Spot checks used the OCR text _text/Ether-And-Relativity-1957-51e9ba.txt and one page image I rendered to the scratchpad (PDF p.7) [read: page image p.7].

Verdict: MINOR. The work is complete and careful. Step 1 read every page, and no central verdict looks likely to change. The defects are small. Some records are out of date. A few points from the paper are missing from the primer. One citation URL disagrees between files, and the verdict labels for two inference steps are open to dispute.

---

1. Coverage

• Every document in the manifest has a digest and an entry in facts.json. Step 0 skipped nothing.
• Stale record (minor). facts.json says text_available: "none" and says pdftotext returned only the bibcode stamp. That was true when Step 1 ran. Step 0 has since OCR'd all 19 pages, and the OCR text is good, except for equations and the junk lines at the top of each page. facts.json should now say text_available: "partial (OCR; equations garbled)" and keep "page images" as the method of reading.

2. Shortcuts

• The task brief says large documents were sampled in Step 1. This project was not sampled. The paper is 19 pages, and all 19 were read from images. The reviewers cite pages from 1 to 19, and their citations match the text (see Section 4).
• There is no deep/ folder. That is acceptable here, because Step 1 read the whole paper from page images, which is at least as thorough as a deep read. The claims that carry the verdicts are the equations (7), (7a), (7b) and (8), the clock footnote on p.16 and the "entails" argument on p.10. All of them were read from the source, and three reviewers checked them independently.
• No central claim rests on a digest alone.

3. Missed material

I found these in the digests and the OCR text. None of them changes a verdict. Items 1 and 2 matter most to the reader.

1. One-way speed of light (pp.16, 18). Builder makes explicit use of the fact that "true one-way measurements of light velocity are impossible (so that we can measure only the average value ... over go-and-return paths)" [read: p.16, OCR line ~1161]. He uses it to explain why every frame measures c and why simultaneity is relative (p.18). This is the key to how Lorentz ether theory (LET) reproduces special relativity (SR): the synchronisation convention. The primer never mentions it, the arbiter does not either, and only the contrarian notes it once. It is the ideal teaching point for this paper. It ties directly into the primer's Section 5 finding that Builder "quietly assumes absolute simultaneity".
2. Einstein's 1905 magnet-and-conductor argument rejected (pp.6-7). Builder's pole and charge cases (6a), (6b) are aimed at the opening of Einstein's 1905 paper. Builder says the "relative motion only" view of induction holds only when retardation is negligible, and "must be rejected" as a fundamental feature [read: page image p.7]. The contrarian mentions this (contrarian.md line 59). Claims.md, the arbiter and the primer do not. The primer's evidence table has no row for (6a)/(6b) at all. A reader who knows Griffiths-level electricity and magnetism (E&M) would gain from it.
3. Einstein's 1920 and 1924 "ether" (pp.14-15). Builder quotes Einstein's Leiden lecture at length ("according to the general theory of relativity space without ether is unthinkable") and Einstein's own contrast with Lorentz's ether [read: p.14-15]. The contrarian covers it in one sentence. The primer leaves it out. The reader profile favours alternative-science readers, who often cite this quote, so the primer should give it in one paragraph with what it does and does not imply. Einstein's ether has no state of motion. The contrarian notes this.
4. Mach's own 1883 ether alternative (p.15). Builder reports that Mach himself considered a space-filling medium "admissible and sufficient" and preferred his own hypothesis only as more "expedient provisionally" [read: p.15, OCR lines ~938, ~976]. This is not recorded anywhere in the review. It bears on Claim 4(ii).
5. Broad (1923) "two independent hypotheses" (p.16 footnote). The physicist and mathematician name Broad as the target of Builder's footnote. The arbiter and primer discuss the argument but never name who made it. This is a small attribution gap.
6. Acknowledged predecessors (p.19). Wiechert, Lodge and Ives are named as earlier proponents of absolute concepts. Ives appears in the review only as Ives-Stilwell (the experiment). Herbert E. Ives's own pro-ether writings of the 1930s-50s are part of the context [memory: medium] and could go into further reading.
7. "Reality" of relativistic effects (p.18). Builder answers "are length contraction and time dilation real?" with "really-observable, caused by motion relative to the ether". This is a standard question students ask, and the primer could answer it in two lines.

4. Accuracy spot checks

example.py was run with .venv/bin/python and exited 0. The symbolic boost minus (7b) gives [0, 0]. The (7a) bracket equals the β² expansion of the exact field. At β = 0.6 the force is 2.8838e-10 N in one frame and 2.3071e-10 N in the other, the net third-law mismatch is 5.7677e-11 N, and the light-clock ratio is 0.8000. Every number in primer Sections 4 and 5 matches the script output.

5. Process

• Source tags. Tags are present and honest throughout. [memory] items carry confidence levels. Weak items are flagged rather than hidden: the Nagel 2015 bound is marked "check failed" and dropped, and the CMB discovery date is "[memory: medium, not checked]". Small gap: the primer could state the CMB dipole discovery date with a [web] source, or drop it. The physicist gives 1965 for the CMB itself. The dipole was measured in the 1970s [memory: medium]. That still falls after 1958, so the claim stands.
• Web checks. The arbiter made 8 checks and logged them in a table, with one failure stated plainly. The reviewers checked Kennedy-Thorndike, LET equivalence, test-theory bounds, the CMB and Gravity Probe B (GP-B). This is adequate.
• Source gap. It is repeated in claims.md, the arbiter and the primer, as required. The gap is the missing Dingle letter and Builder's 1957a and 1957b papers. The arbiter correctly records that Builder's paraphrase of Dingle and the web summary of Dingle's later position differ.
• Verdict scale. It is applied consistently, and the arbiter states its rules. One judgment call should be recorded rather than changed silently. "Internally inconsistent" is used for Claim 1 ("demands") and Claim 4(i) ("entails"). Both are better described as non sequiturs: the conclusion does not follow from the premises, but it does not contradict them. Builder's absolute frame S0 is itself one of the frames in which the equations have the same form, so it is arguable that no self-contradiction exists. The arbiter's justification (frame-independence premise against a privileged frame, and Newton's precedent of absolute acceleration without absolute velocity) is reasonable but disputable. A note in the arbiter would make the call visible to the reader. Verdicts need not change.
• Primer against STEP 4 and the reader profile. All 9 sections are present and in order, with answers at the end. Symbols are defined (γ, β, θ, k, Gaussian units, proper time τ). Acronyms are defined: SR, LET, CMB, KT and GP-B in the arbiter. Real numbers are worked through. Prerequisites name textbooks, without chapter numbers except where the reviewers were sure. No tensors or four-vectors are assumed. Good.
• example.py runs and reproduces the primer.

6. Remediation actions

Deep reads (all small; the paper was already read in full from images)

File changes

• facts.json: set text_available to "partial (OCR text exists; equations garbled)". Add to key_claims the 1905 magnet-and-conductor rejection and the one-way light-speed point. Name Broad (1923) and Mach's 1883 medium remark in the summary.
• claims.md: correct the page references (clock-slowing footnote at p.16, pedagogy at p.18; claim 4's Mach material at pp.14-15). No change to the claims themselves.
• arbiter.md: add a "QE revision" section that covers four items. (a) Replace the Dingle URL with https://www.nature.com/articles/179865a0 or mark 1791071a0 as unverified. (b) Reword "seems impractical" as described in Section 4, check 6. (c) Note under Claims 1 and 4(i) that "Internally inconsistent" is a judgment call for what is arguably a non sequitur, and give the alternative label. (d) Add a short ruling on (6a)/(6b), which is expected to follow Claim 3. No change to the headline verdicts is expected.
• primer.md: make four additions. (a) A short subsection, in Section 3 or 4, on one-way light speed and the synchronisation convention, linked to the "absolute simultaneity" slip in Section 5. (b) One paragraph on Einstein's 1905 magnet-and-conductor example and Builder's rejection of it, plus a row for (6a)/(6b) in the evidence table. (c) One paragraph on Einstein's 1920 "ether" quote and why it does not support a Lorentz ether, with Einstein's own words as Builder quotes them. (d) Name Broad (1923) as the source of the "two hypotheses" objection. Optionally, add a check-yourself question on one-way light speed.
• example.py: no change needed.

Recheck

Recheck of the QE remediation list. Files re-read: facts.json, claims.md, arbiter.md (including its "QE revision" section), deep/Ether-And-Relativity-1957-51e9ba.md, primer.md; example.py re-run with the venv (exit 0, numbers unchanged).

Deep reads

Defects and file changes

The arbiter and facts are now consistent with the QE list. The primer was not updated, so the reader's document still lacks the four additions.

Final: OPEN (4 issues)

Recheck 2

Re-read primer.md, which was updated after the first recheck. The four items marked as remaining were checked by grep and by reading the new text.

Final: PASS