Mills-Randal: Quality engineering (STEP 4b)

QE agent (Opus), 2026-10-09. Project: R. L. Mills, The Grand Unified Theory of Classical Physics (GUT-CP), one PDF (1,876 pp, "December 10, 2018 Edition", file name "2020 Ed Web"); slug GUT-CP-2020-Ed-Web-b290db. Tags: [read: p.N] = the extracted text _text/GUT-CP-2020-Ed-Web-b290db.txt at that PDF page, or the named research file; [calc] = recomputed by me; [memory] with confidence.

Verdict: MINOR

The review of claims 1-5 is solid, well sourced and fair, and its verdicts are unlikely to change. The defects are (1) three digest-only areas that carry part of a central claim (claim 7 entirely; Ch.11 for claims 2 and 4; the general-relativity half of claim 6), (2) one falsified prediction the deep reader found and every later file dropped (the book's statement that gravitational waves do not exist, kept in a 2018-dated edition after LIGO's 2015 detection), plus a few other book claims nobody carried forward, and (3) corrections the arbiter ordered that have not yet reached claims.md, physicist.md and sites.md. All of these can be fixed with deep reads, a "QE revision" section in arbiter.md and primer edits. The claim 7 verdict is provisional and may change after the Ch.24-30 deep read. MINOR allows that change through the QE revision section, so Steps 2-4 do not need to be redone.

What I did (my own method, stated plainly)

• Read prompt.txt, manifest.json, claims.md, facts.json (index and per-part entries), all three reviews, arbiter.md, primer.md, sites.md, and the research notes on patents, the press, Wikipedia, Rowan and FindLaw. Listed every section heading of the deep-read file (4,944 lines) and grepped it for specific items.
• Digests. The 130 digests add up to 1.75 MB (about 450k tokens), more than one context can hold alongside the other inputs. So I did not read every digest line. Instead I (a) checked that all 130 parts exist and cover pp.1-1876 with no gap; (b) took every sentence across all 130 digests that mentions evidence or history terms (replication, laboratories, institutions, patents, the company, calorimetry, EPR, NMR, Raman, XPS, SunCell, named validators); (c) took every "calculated vs experimental / observed / predicted" sentence in the ranges no one deep-read (pp.201-247, 333-350, 445-504, 1521-1620, 1666-1679); (d) ran targeted greps on Ch.17-21, Ch.35 and the Introduction; and (e) checked suspected misses against the extracted book text. Material in digests I did not reach this way may still be missed. This is a disclosed QE shortcut.
• Ran example.py (it runs and its output matches primer section 5, with one small discrepancy noted below). Spot-checked 8 load-bearing facts against the book text and the Investigator's downloaded PDFs (table below).

1. Coverage

• Manifest: 1 document, 1,876 pages, 50 OCR pages, 130 chunks. Digests: 130 of 130 present. Their page ranges join without a gap (pp.1-11 ... pp.1867-1876). Nothing was skipped. Pass.
• facts.json has one entry for the book plus 15 per-part entries that cover every chapter and appendix. Each how_read field describes what was actually done. Pass.
• Note: digest parts 70-77 cover 30-50 pages each (for example part 71 covers pp.796-836), against about 10 pages for most parts. These are the table-heavy functional-group pages of Ch.15 (pdf 801-1082), the area where claim 4's "were the parameters fitted?" question sits. That density is probably below the "1-3 sentences per 2-3 paragraphs" rule. It is only a defect because no deep read covers those pages (item S5).

2. Shortcuts (central material resting on digests alone)

3. Missed material (in the sources or deep reads, absent from facts.json, claims.md, the reviews, arbiter and primer)

M1. Gravitational waves (most important miss). The book says twice that gravity cannot form a transverse light-speed wave and that this is "consistent with the absence of the direct experimental observation of gravity waves" [read: p.1642, refs 128-129 = LIGO 2009, Shawhan 2010]. It also says: "The time dependent gravitational field fluctuations would only be experienced radially in the near field with no transverse time-dependent gravity wave effect in the far field consistent with the absence of the observation of gravity waves [32-33]" [read: p.1678]. LIGO detected GW150914 in September 2015 (announced February 2016) [memory, high], and GW170817 (2017) arrived together with light, showing gravitational waves travel at c [memory, high]. The project copy is dated December 2018 and still prints the claim. The Ch.32 deep reader recorded this [read: deep file line 4042]; facts.json, claims.md, all three reviews, the arbiter and the primer dropped it. It is a clear test of claim 6 and it went against the book. It belongs in the claim 6 verdict as Tested, not supported, after the arbiter checks the LIGO facts on the web. The 2025 edition (research/files/GUT-CP-online-2026-04-28.pdf) should be searched to see whether the passage changed.

M2. The Higgs prediction before 2012, and its later reinterpretation. "Such a mass conveying particle will not be observed in experiments performed at ... Fermilab, the Large Hadron Collider (LHC), or any future collider [31]" (ref 31 = Cho, Science 2010). The same paragraph then reinterprets the CERN 126 GeV result as "a high-energy neutron resonance" [read: p.1678; also p.1720]. The arbiter has only the mass comparison (128.75 vs 125.20 GeV, +2.8 percent). The record should also carry the prediction that the particle would not be observed, and that the reading changed after the fact. This bears on how the "predicted in 1995" claims are weighed.

M3. Fifth force, pseudoelectrons and the gravitational mass of the free electron (Ch.35). The book claims a free electron has zero gravitational mass. It cites the Witteborn and Fairbank free-fall result (PRL 19, 1049, 1967: "less than 0.09 m_e"), rejects the Schiff-Barnhill shielding explanation, and quotes a 2013 CERN Courier remark that electron drops gave g "consistent with zero" [read: p.1680, 1686, 1702]. It goes on to claim that "negative gravity (fifth force) can be created by forcing matter into negative curvature" (pseudoelectrons), with a threshold photon energy of 44.2 eV [read: digest p.1684, 1698-1699]. The Investigator found the matching patents: "antigravitational force" (1989, 1994) and "Fifth-force apparatus and method for propulsion" (2007), plus a 2025 "Reactionless propulsion" paper [read: research/2026-10-09-google-patents.md]. None of this is in claims.md or any review. It is a distinct, testable claim with a real experiment behind it (Witteborn-Fairbank), and a contrarian reader would find it interesting. It should be added to claim 6 (or as a sub-claim) and reviewed in the QE revision.

M4. Hydrino-catalysed fusion (HCF). pp.272-274 propose that deuterium or tritium hydrinos at high p fuse more readily, and cite hydrated-silver-shot detonations with power densities "comparable to those of nuclear weapons [50-54]" [read: digest p.272-274; the deep read for pp.265-286 covers these pages]. It appears once in facts.json and nowhere in claims.md, the reviews, the arbiter or the primer. It links Mills to the cold-fusion literature (the NASA TM itself names "the subsequent fusing of the 'shrunken' atoms" as one hypothesis [read: NASA TM p.1]). It should be recorded as a claim of the book, with the observable it predicts (neutrons, tritium, 3He).

M5. The inventor's own count of confirmations. "The existence of such states has been confirmed by the data presented in over 100 published journal articles and over 50 independent test reports and articles [22]", where ref 22 is the company URL [read: p.37]. This is the interested party's own characterization, and it belongs in the absence-of-evidence analysis next to the reviewers' finding that the test reports found are company-hosted and authored by hired observers or co-authors. Neither statement should be given without the other.

M6. The NASA TM nuance (for fairness). The NASA authors also wrote that explaining all the excess by recombination "requires a recombination efficiency of quite a high magnitude" (above 0.55, rising to about 1 at 5 A), which the data "marginally admit" [read: research/files/NASA-TM-Niedra-1996, p.4]. They add that the loaned cell's maker documents "50 W steady excess heat for a continuous period exceeding hundreds of days" (the company's claim, quoted by NASA) [read: p.1]. The arbiter and primer give only "recombination could not be excluded". The fuller statement leans slightly toward the claim, and SOURCE HONESTY requires it.

M7. The "fast H" / Balmer-broadening evidence class. The book cites seven papers (refs 36-42, 2005-2010) on fast hydrogen atoms and broadened Balmer lines, and returns to them in Ch.35 ("observed experimentally [19-25]") [read: digest p.85, p.1689]. The Investigator found related outside work (Phillips 2008, a Mills co-author; Loureiro and Amorim 2010, who argue against). The physicist set it aside as [memory, low-medium]. The arbiter gives it one line and the primer nothing. It is a distinct experimental anomaly class and should be weighed in claim 2 (deep read or web check of Loureiro-Amorim).

M8. Helium excited states (Ch.9). The book claims more than 100 He I levels with r^2 = 0.999994 and an average relative difference of about 2e-4, and calls this strong support for hydrinos [read: p.345-346]. Example rows: 1s2s 3S -4.78116 vs NIST -4.76777 eV (0.28 percent); high-n rows agree to about 1e-4 [read: digest p.346-349]. No review mentions it. A reviewer should note that high-n levels approach the hydrogenic -13.6/n^2 eV for any model with a screened core, so agreement there is cheap; the n = 2 rows are the real test.

M9. Other book claims nobody carried forward (lower priority): the KEK "experimental confirmation" of the free-electron charge distribution [read: digest p.205]; the N2 orbital tomography offered as "direct evidence" that electrons are 2D shells [read: digest p.489]; the liquefaction temperature of H2(1/p) predicted "significantly higher" than helium's [read: digest p.480]; Steinhardt and Spergel cited as saying hydrino-like properties are what dark matter needs [read: p.1634, p.1642] (a third party's statement used by an interested party, to be checked against the original); "absolute space confirmed experimentally" from quasar and supernova time dilation [read: digest p.1678]; the Mossbauer, MRI and pattern-recognition patents the book cites [read: digest p.189, 1550, 1803].

M10. Investigator historical record left out of the arbiter and primer (pertinent, in sites.md and research/): - Scientific American 2014: Mills said hydrinos had not been isolated [read: research/2026-10-09-wikipedia.md, a tertiary pointer; original not read]. This sits against the book's "hydrinos are isolated and confirmed" [read: digest p.83]. - The Group Director's declaration that the patent "appeared to involve cold fusion and perpetual motion" [read: research/pages/wiki-blp.txt, citing Rimmer 2011]. Rimmer's 2011 article and Bob Park's 2002 account of the "outside inquiry" are named leads. - The Harvard Crimson (2000): "over $25 million" from investors; a board with Shelby Brewer (former NRC official), Vice Admiral Kalleres and Aris Melissaratos (Westinghouse STC); Mills's claim of Navy research trials; critics Georgi and Chu; the 2000 law-firm letters to four physicists [read: research/2026-10-09-press-crimson-crossref.md, -wikipedia.md]. - The Jansson 1997 Rowan thesis, done under a HydroCatalysis service contract (the earliest academic calorimetry for the company), and the Rowan 2008-09 tests [read: research/2026-10-09-rowan-digital-works.md]. - Mills's 115-family patent record, including pre-hydrino medical patents and the 1989 "Energy emission process" [read: research/2026-10-09-google-patents.md]. - NJEDA grant of about $1.1M (2013) [read: research/2026-10-09-wikipedia.md]. - Hagen-Mills 2024/26 broad-line EPR, Selke 2025 and Phillips/Loureiro-Amorim: in the arbiter, missing from the primer's experimental record.

4. Accuracy spot checks

No load-bearing number checked was wrong.

5. Retention (pertinent material left out because judged negatively)

• I found no case where the arbiter or primer dropped evidence for Mills because a reviewer judged it negatively. The NASA, Thermacore, Glumac, Hagen, NIAC and IIT Kanpur items are all kept and stated fairly, with stakes marked. The contrarian's "strongest honest case" points and the mathematician's look-elsewhere count are kept. Pass on the main test.
• Smaller retention gaps:
• The contrarian's note that Maris found standard bubble models fail on the helium mobility peaks: dropped by the arbiter and primer. Restore it in primer row 5d.
• The NASA "high recombination efficiency" nuance (M6).
• The fast-H evidence class (M7): the physicist set it aside for lack of sources, and it then fell out of the primer entirely.
• The primer keeps none of the Investigator's speculative relatives (Zhigalov, Nevolin, Baranov-Zatelepin) and none of the patent record. The arbiter keeps the relatives, so this is acceptable for a study document, but the patent record should appear once (M10).
• Against-evidence also counts under this rule and was dropped: the gravitational-wave failure (M1), the Higgs non-observation prediction (M2) and Scientific American 2014 (M10).

6. Process

• Source tags. Present and mostly honest throughout. Exceptions:
• The physicist's web check 2 tagged "[2008] EWHC 2763 ... confirmed as dismissed appeal" as [web] from a search summary. It was wrong, and the arbiter caught and corrected it.
• The primer tags Mills's background "[memory, medium]" when the book states it: "MD from Harvard Medical School ... BA in chemistry ... Franklin & Marshall ... studying electrical engineering at MIT" [read: p.1875, also p.439, 1518]. Retag it as [read], and note that it is the publisher's description.
• Web checks. The arbiter used 6 of 8 calls, all on verdict-bearing items. The physicist and contrarian each used about 6. Items still at [memory] (Bullet Cluster, Benson accuracy, Tonomura, BICEP2 withdrawal) are marked as not decisive. The new LIGO, Witteborn-Fairbank and Higgs items (M1-M3) need web checks in the QE revision.
• Verdict scale.
• Applied consistently, and the arbiter reconciles the reviewers' different labels explicitly.
• The arbiter uses "Established as arithmetic" (a qualified Established) and "not assessed". Both are acceptable as stated qualifiers, but "not assessed" is not a scale verdict. Once S3 is deep-read, the GR part needs a scale verdict.
• Primer.
• Follows the nine STEP 4 sections in order, with a worked numerical example, an evidence table, check-yourself questions with answers, and further reading. It is written for the reader profile, with symbols defined.
• Undefined acronyms in sections 6 and 9: SBIR (Small Business Innovation Research), ESCA, XRISM, SH0ES, BICEP/Keck, PDG, ToF-SIMS, CODATA (in the tables). Define each on first use.
• Small numerical slip, section 5 step 5: "through order 2: 1.6 x 10^-8 too low". example.py prints -1.475e-8 (1.5 x 10^-8). Fix the text.
• example.py. Runs cleanly in the project venv; every other printed number matches the primer.
• Corrections pipeline (not yet applied).
• physicist.md still says the UK appeal was "dismissed" (section 3 historical record; section 11 item 2).
• sites.md Leads 4 still says the appeal was "not yet read".
• claims.md still has "p up to 137" without the 2025 value of 126, "replicated by Bruker" without the qualifiers, and the bare "no fitted parameters".
• facts.json has no Hitomi/XRISM note for the 3.48 keV item.
• These are arbiter corrections 1-4 and 7. The orchestrator should apply them before STEP 5.

7. Remediation (one round)

Deep reads (requested below, most important first): Ch.24-30; Ch.11; Ch.31-34 including pp.1642 and 1678; Ch.35; Ch.9; Ch.15 middle; Introduction; Ch.3-4.

facts.json. - Add to the Ch.31-35 entry: the gravitational-wave statements (p.1642, p.1678), the Higgs "will not be observed" statement (p.1678), the Witteborn-Fairbank citation and the fifth-force/pseudoelectron claim (pp.1680-1702). - Add the HCF fusion proposal to the Ch.5-6 entry. - Add the "over 50 independent test reports" statement to the front-matter entry (p.37). - Apply arbiter correction 7 (3.48 keV: Hitomi/XRISM).

claims.md. - Apply arbiter corrections 2-4. - Add to claim 6: "gravity has no transverse far-field waves" (p.1642, p.1678); "the Higgs boson will not be observed" (p.1678); "a free electron has zero gravitational mass, and negatively curved 'pseudoelectrons' produce a repulsive fifth force" (Ch.35). - Add to claim 2: hydrino-catalysed fusion (pp.272-274). - Update the "Orders for further deep reads" list after the deep reads run.

arbiter.md, "QE revision" section. - (a) Claim 7 verdict after the Ch.24-30 deep read. Positronium g/2 1.001159652137 vs 1.001159652188(4) is a precision comparison like the electron g and should get a scale verdict. - (b) Claim 6: the gravitational-wave statement, after web-checking GW150914 and GW170817 (likely Tested, not supported); the Higgs non-observation prediction and its reinterpretation; the GR half (perihelion Table 32.1, deflection Table 32.2) once deep-read; the Witteborn-Fairbank fifth-force sub-claim (web-check the Lockhart-Witteborn-Fairbank follow-up and the mainstream status of the Schiff-Barnhill effect). - (c) Claim 2: the fast-H evidence class, the HCF proposal and the NASA recombination-efficiency nuance. - (d) Claim 4: the Ch.11 H2 derivation (whether any input is fitted) and the Ch.15 parameter-selection read. - (e) Absence of evidence: the book's "over 50 independent test reports" beside the company-hosted record; Scientific American 2014; the Kepplinger "cold fusion and perpetual motion" remark; the Crimson 2000 Navy-trials claim; the Jansson 1997 service-contract thesis. - (f) Restore the full USPTO quote ("...before issuance of the Notice").

primer.md. - Fix the step 5 number (1.5 x 10^-8). - Retag the biography as [read: p.1875]. - Define the acronyms listed in section 6. - In the evidence table, add rows for gravitational waves and the Higgs prediction under 6, and for the free-electron gravity / fifth force; update row 7 after the deep read. - In "experimental record", add the NASA efficiency nuance, the fast-H anomaly class, Hagen-Mills 2024/26 broad-line EPR, Scientific American 2014 and one line on the patent record. Add the Maris point to row 5d. - Add the gravitational-wave case to "What is wrong and why the mistake is instructive": a prediction kept in print after the decisive observation.

physicist.md / sites.md. Apply arbiter correction 1 ("allowed and remitted"); update sites.md Leads 4.

8. Corrections to earlier documents (for the orchestrator)

1. facts.json, claims.md, the three reviews, arbiter.md, primer.md: the gravitational-wave claim was dropped. - The Ch.32 deep read [deep file line 4042] records the book's statement [read: p.1642] that gravity waves do not exist (in a December 2018 edition, after GW150914), and p.1678 repeats it. - No later file carries it. It is a decisive observational test of claim 6 and goes against the book. - Add it as described in section 7.
2. primer.md section 5, step 5: "through order 2: 1.6 x 10^-8 too low" should read "1.5 x 10^-8 too low". Evidence: example.py, Part E prints -1.475e-8.
3. primer.md section 1: "[memory, medium]" on Mills's training is the book's own statement [read: p.1875, p.439, p.1518]. Retag it as [read], the publisher's description.
4. arbiter.md (Investigator's findings), primer.md (patent offices), physicist.md section 3: the quote "no one involved in the withdrawal had reviewed the application before" is cut short. The FindLaw opinion reads "undisputed that no one involved in the withdrawal had reviewed the '294 patent application before issuance of the Notice" [read: research/2026-10-09-findlaw.md].
5. arbiter.md, primer.md (NASA TM-107167): add the authors' statement that recombination would need "quite a high" efficiency (above 0.55, rising to about 1 at 5 A), which the data "marginally admit" [read: NASA TM p.4]. "Could not exclude" alone understates the authors' own weighing.
6. Arbiter corrections 1-4 and 7 are not yet applied to physicist.md ("dismissed"), sites.md ("appeal not yet read"), claims.md (p <= 126, Bruker qualifiers, "no fitted parameters") and facts.json (3.48 keV). Apply them, then search work/rankings/entries for "dismissed" before the entry is written.

Deep-read requests (most important first)

1. GUT-CP-2020-Ed-Web-b290db, pdf 1521-1580, Opus: Ch.24-30, claim 7.
2. GUT-CP-2020-Ed-Web-b290db, pdf 445-504, Opus: Ch.11 (H2, H2+, dihydrino).
3. GUT-CP-2020-Ed-Web-b290db, pdf 1581-1620 and 1666-1679, Opus: Ch.31, first half of Ch.32, Ch.34 (GR, perihelion, deflection, gravitational waves, Higgs, absolute space).
4. GUT-CP-2020-Ed-Web-b290db, pdf 1680-1703, Sonnet: Ch.35 fifth force, pseudoelectrons, Witteborn-Fairbank.
5. GUT-CP-2020-Ed-Web-b290db, pdf 333-350, Sonnet: Ch.9 helium excited states.
6. GUT-CP-2020-Ed-Web-b290db, pdf 801-1082, Sonnet: Ch.15 parameter selection.
7. GUT-CP-2020-Ed-Web-b290db, pdf 1-86, Sonnet: Introduction, the inventor's evidence claims.
8. GUT-CP-2020-Ed-Web-b290db, pdf 201-247, Sonnet: Ch.3-4 free electron and photon.

The question for each is in the structured deep-read list returned with this file.

Recheck

QE recheck (STEP 4b), 2026-10-09. I re-read the current claims.md, arbiter.md, primer.md, facts.json, physicist.md and sites.md against the defects above, using grep on each file. I did not re-open the book text or the deep-read files.

Fixed - S1-S6 and S7 (digest-only areas): arbiter.md now has a "QE revision" section built on deep reads of Ch.24-30, Ch.11, Ch.31-32, Ch.34, Ch.35, Ch.9, Ch.15 (prose) and the Introduction (and Ch.3-4). Claim 7 is no longer provisional, and the GR half of claim 6 now has scale verdicts. The ranges still not deep-read are stated openly (Ch.17-21, Ch.23, Ch.33, the Ch.15 numerical tables), which is acceptable. S8 and S9 were already acceptable or low. - M1 gravitational waves, M2 Higgs: carried in arbiter.md, claims.md, facts.json and primer.md ("tested, not supported"; GW170817 speed noted). - M3 fifth force and Witteborn-Fairbank, M4 hydrino-catalysed fusion, M5 "over 50 independent test reports", M7 fast-H, M8 helium (Ch.9), M9 (KEK and others): present in arbiter.md, with most also in claims.md, facts.json and the primer. - M6 NASA recombination efficiency: in arbiter.md, claims.md and the primer. - M10 (Scientific American 2014, patent record, Hagen-Mills EPR): in arbiter.md and the primer. - Maris point in primer row 5d: restored. - USPTO quote ("...before issuance of the Notice"): restored in arbiter.md and primer.md. physicist.md line 70 has the full phrase. - primer.md step 5 number: now 1.5 x 10^-8. Mills biography: retagged [read: p.1875]. Acronyms (SBIR, ESCA, XRISM, SH0ES, BICEP/Keck, CODATA): defined on first use. PDG and ToF-SIMS no longer appear. - claims.md: p <= 126, Bruker qualifiers and the qualified "no fitted parameters" applied. facts.json: the Hitomi/XRISM note for 3.48 keV is present. - The positronium g/2 mistake in the arbiter and contrarian was caught in the deep read and corrected (it is the free-electron value re-quoted from Ch.1).

Remaining (2) 1. physicist.md still says the UK appeal [2008] EWHC 2763 (Pat) "was dismissed" (line 70, section 3 historical record) and "confirmed as dismissed appeal" (line 158, section 11 item 2). The correct statement is "allowed and remitted". claims.md says this was "applied by the orchestrator", but physicist.md does not show it. arbiter.md line 456 also lists it as still to be applied. 2. sites.md Leads 4 (line 97) still says "appeal [2008] EWHC 2763 (Pat) not yet read". The appeal was read and found allowed and remitted.

Both are stale-copy defects of the kind the project corrections rule covers. The orchestrator should fix them, then search work/rankings/entries for "dismissed" before the entry is written. No verdict is affected.

Final: OPEN (2 issues)

Orchestrator fix (2026-10-09)

Orchestrator fix (2026-10-09): physicist.md lines 70 and 158 "dismissed" -> "allowed and remitted"; sites.md line 97 "not yet read" -> "since read: allowed and remitted"; arbiter.md item 11 marked applied.

Final: PASS