Kelly-Patrick: Quality Engineering (Step 4b)

QE run 2026-10-09. Inputs read: _text/manifest.json; the digests in digest/ (see "How QE read the digests" below); deep/PJKbook-9cc3a5.md (by section); facts.json, claims.md, sites.md, the research/ notes (via sites.md "Leads for review"), physicist.md, mathematician.md, contrarian.md, arbiter.md and primer.md. I ran example.py. Spot checks were made against _text/PJKbook-9cc3a5.txt. Page numbers are PDF page numbers (the === page N === markers), as in the rest of the project.

How QE read the digests. The project has one document (PJKbook.pdf, V32.9, 2,481 pages) in 103 digest chunks. I read every digest line in full for every range that was not deep-read: chunks 01-09 (pp.1-222), 43-47 (pp.999-1106), 50-54 (pp.1147-1285), 58-60 (pp.1369-1453), 71 (pp.1686-1716), 80-97 (pp.1882-2333) and 102-103 (pp.2433-2481). For the deep-read ranges (chunks 10-42, 48-49, 55-57, 61-70, 72-79, 98-101) the deep reads replace the digests. There I checked the digests mechanically rather than line by line: I listed every proper name in those digests that does not occur in the deep reads, facts.json or the arbiter, and followed up the ones that were not noise (for example the Caggiano FIVS story, p.888-889, and the climate essay, pp.993-997; both turned out to be in the deep read). This is stated so the recheck knows what QE did and did not do.

Verdict: MINOR

The reviews, the arbiter and the primer are careful, and the arithmetic I rechecked is right. The headline verdicts will not change. The weak point is upstream. The fact-finder sampled the digests for about 40 percent of the book: "about half of the digest lines" for Chapters 1-3a, only one chunk of Chapter 12, about one line in five for Chapter 14, one line in three for Chapter 20, and appendix chunks 60-103 "not read in full" (facts.json how_read fields). Material that was never sampled therefore never reached claims.md or the reviewers. It includes a whole class of claims (motors driven by permanent magnets alone, Chapter 1), several of the book's largest numerical claims (Adams motor "COP 1,223", 27.6 W in and 33.78 kW out; Barbosa and Leal "169 kW", COP about 100; Heins patent test "COP 39.6"), Kelly's own bench claim that a DC-DC converter gave "COP = 1.5", the Papp patent itself (the source of the noble-gas fusion mechanism), the Puharich, Hasebe, Chambers and Spiros electrolysis claims with their numbers, the high-mpg carburettor patents, and the book's medical claims (Chapter 20, Beck).

All of this fits the existing claims 2-7 and the existing lines of reasoning (energy per event times rate; the Gibbs floor; closed-cycle work of conservative forces; chemical-energy bounds). None of it is likely to overturn a verdict, so a "QE revision" section in arbiter.md, with updates to facts.json, claims.md and the primer, is enough. Steps 2-4 do not need to be redone. One fact in facts.json is misread (Adams, below) and must be corrected.

1. Coverage

• All 103 chunks have a digest file. Nothing was skipped by Step 0, and the folder holds no junk files [read: manifest.json, _textprep.log].
• The contents list names Chapter 22 "Dealing With Chemtrails" and Chapter 23 "The Mandela Effect". In the body, Chapter 21 is followed by "Chapter 22: The Sabourin Generator" (text line 53689); the two contents-list chapters are not in this edition's body [read: extracted text lines 479-488, 53461, 53689]. claims.md is correct. It would be clearer to add that the body's Chapter 22 is the Sabourin generator (listed as Chapter 24 in the contents).

2. Shortcuts

1. Digest sampling by the fact-finder (defect). STEP 1b requires the fact-finder to read all digests. facts.json records that it did not (entries #ch1-2, #ch3a, #ch12, #ch14, #ch20, #appendixA). The deep reads cover the central chapters well, so the cost falls on the chapters and appendix items listed in section 3. QE has now read those digests, and section 3 lists what they hold.
2. Central claims resting on a digest alone (defect, partly). Three things rest on digest-level text, or on nothing: - Papp's mechanism. Claim 6 and the arbiter's Papp verdict use the Chapter 8 deep read (p.642), but the Papp patent US 4,428,193 (pp.2196-2228) was digest-only. That patent is the only place where the mechanism is stated: about 36% He, 26% Ne, 17% Ar, 13% Kr and 8% Xe; electrodes holding thorium-232; a "minute fusion reaction" at "100,000,000 degrees F"; a claimed 1,000 hours at 1,200 rpm on one charge [read: digest 91-93, pp.2196-2228]. The arbiter's statement that "plasma recombination returns at most the ionisation energy" [memory, high] does not address the fusion claim the patent actually makes. - Water-splitting appendix (claim 5). The Puharich article and patent, Hasebe, Chambers, Garrett, Spiros, Eccles, Horvath and Aguero (pp.1901-2037) were digest-only. The arbiter says so in open question 10. These texts give numbers that can be checked directly (section 3). - Heins, Kunel, Tseung frame, Coler, Plauson, Rayburn and the Chapter 14 devices. The arbiter says openly that these verdicts are "weaker". That is acceptable for peripheral devices, and the arbiter flags it.
3. Images not rendered. Circuit diagrams, patent figures and most tables were not rendered (arbiter source note). This was acceptable for verdicts based on energy bounds. It is not acceptable for the Spiros data tables (pp.2018-2022, OCR garbled) or Puharich's equations (pp.1909-1912, 1929-1930) if those are to be judged. The remediation deep reads below include page images.
4. Later editions not checked. V34.5 and V34.8 are in research/files/ and have not been compared with V32.9 (arbiter open question 7). This is acceptable as an open item. It is listed for the Investigator, not as a defect.

3. Missed material (from the digests; not in claims.md, the reviews, the arbiter or the primer unless noted)

Kept in the record whatever its likely verdict (SOURCE HONESTY). Each item gives what the book says and where it says it.

A. Permanent-magnet-only motors (Chapter 1). A claim class missing from claims.md. - Kelly: a magnet "does not actually exert any power"; the power "flows from the environment", and the magnetising pulse "polarises the quantum environment" [read: digest 03, p.42]. - ShenHe Wang (China): a 5 kW colloidal-magnet generator. "Two of Wang's 5 kW generators completed China's mandatory six-month 'Reliability and Safety' testing in April 2008", and a Chinese consortium was said to be buying coal-fired stations to refit with them [read: digest 03, pp.42-43]. This is a claim of an official test, with no source given. It is a lead for the Investigator. - Howard Johnson, US 4,151,431 (1979): a granted patent for "a motor powered solely by permanent magnets", with a claimed "mechanical advantage greater than 100 to 1" [read: digest 03 p.49; digest 60 pp.1437-1444]. Also covered: Ecklin-Brown US 3,879,622; Tracy, Teal, Jines US 3,469,130; Kundel US 7,151,332; Hohl's V-track (the Ascanius replication "did work"); Soukup; Donald Kelly US 4,179,633; Diduck US 2007/0296284 [read: digests 03-04, 88, 90]. - Perendev (Mike Brady): "not less than 100 kilowatts". Kelly himself says this "has not been confirmed", that the only test was "a brief test by Sterling Allan", and that by mid-2010 Brady's backers were "most unhappy" [read: digest 04 pp.78-79]. Kelly's caveat should be kept. - Kelly's own warning: repulsion-mode magnets reportedly "lost their magnetisation" after about three months [read: digest 04 p.82]. - Needs a verdict. For rigid magnets with fixed magnetisation the magnetostatic force is conservative, so any closed cycle gives zero net work [memory, high]. This is the same argument the arbiter uses for gravity in claim 4. Gating, shielding or moving stator magnets by an external drive (Kundel, Kelly, Flynn) moves the question onto that drive's energy. Suggested placement: a new sub-verdict under claim 3 ("magnets alone drive continuous rotation": Refuted by calculation for a closed cycle; the Wang test and Perendev output Never properly tested).

B. Large numerical claims from Chapters 2-3a never reviewed. - Adams motor: a builder "Maimariati" with "a Coefficient Of Performance of 1,223 ... His input power is 27.6 watts and the output power is 33.78 kilowatts" [read: extracted text p.93, spot-checked]. Also p.29: one replication with "27 W" input and "32 kW" output; Adams "typically, at least 800% efficient" [read: digest 02 pp.28-29; digest 04 p.90]. Adams was reportedly told he would be killed if he shared the information (p.28; suppression episode). - RotoVerter: "at least 10 times input power as output", said to have been reproduced by "several independent researchers"; Muller motor "35 kilowatt"; coil-shorting "about 100 times" boost [read: digest 05-06, pp.118-123]. - Tewari (US 2015/0084467): COP 2-3 from "reduced reaction". Kelly notes Tewari does not claim self-powering [read: digest 05 pp.102-103]. This is a fair caveat to keep. - Tseung magnetic frame: one replicator got "COP = 1.5" after reducing three magnets to two [read: digest 07 p.149]. - Thane Heins CA 2594905: patent test figures 0.29 W in, 11.51 W out, "a COP of 39.6" [read: extracted text p.150, spot-checked]. facts.json gives only 3.77 and "thirty times". - Klingelhofer "Gabriel Device": 60 W in, 480 W out, "COP = 8" [read: digest 08 p.170]. - Markov EP 0844626: output emf rises "dozens of times" with a single primary; transformer 20-30 times lighter [read: digest 07 pp.152-164]. - Figuera (d.1908): a 1902 Daily Mail report of 550 V and a 20 hp motor driven from "a small apparatus" [read: digest 08 p.170]. Not checked; a newspaper lead. - Barbosa and Leal (WO 2013/104041-43): "COP of about 100 as a minimum"; a video of "22 W producing 6 kW"; "169 kW flowing from the ground ... output 104 times the input" (p.37 table: "169 kW, COP=102.4"). A forum builder "Clarence" reports 3 kW for his household from many ground rods [read: digest 08-09, pp.179-203; digest 03 p.37]. - Matchett / Ascanius "static" rods on the mains live wire, with measurements (2.6 V with the fuse out; 129-162 V with the fuse in; 61 V under a 7 W LED load). Kelly doubts that these mains circuits give net power ("meters bill average voltage times average current") and does not recommend them [read: digest 09 pp.205-211]. Kelly's doubt is pertinent and should be kept. - Annis/Eberly appendix text, quoted on p.219: a narrow band of rise time gives "COPs in the range of 200 to 400 or greater", with COP "Output power/Drive Power" for the switches, without the coupling circuit. The arbiter already has this right.

C. Kelly's own measurements and caveats (pertinent; partly missing). - p.1259: Kelly's own bench claim that a Chinese LM2587 DC-DC converter gave "COP = 1.5": input 800 mA at 3.33 V (2.664 W), output 389 mA at 10.3 V (4.01 W) [read: digest 54 p.1259]. This is the only COP above 1 in the book that Kelly says he measured himself. It bears directly on the primer's lesson that "peaks and products are not power": a switching converter's input current is pulsed, so a DC ammeter's reading of it is unreliable. It belongs beside the Joule Thief lux test in claim 2. - pp.1089-1091 (Chapter 12): Kelly's own explanation that volts times amps is not power for non-sine or reactive loads, with a worked example in which a reading of "COP = 1.56" "may be wrong", and his statement that true power needs sampling wattmeters [read: digest 47 pp.1089-1091]. This is Kelly's strongest caveat and the primer should quote it. The reviews cite p.314 and p.544 but not this passage. - pp.1369-1373 (Chapter 19, deep-read): a South African builder's pulse generator, 0.89 W in and "40 watts or close to it", "COP of 45", three weeks running, later 60 W and 150 W versions [read: facts.json ch18-19]. In facts.json, but it has no verdict in the arbiter. It belongs in the claim 2 "Never properly tested" list. - p.1268: refrigerators described as "about 300% efficient"; p.1163: Perkins shear heater "at least 100% efficient", "measurement methods ... not discussed"; p.1217: the ram pump "COP of infinity", with Kelly asking whether rejecting compressed-air self-running engines is "bias" [read: digests 52-54]. These are useful teaching cases for the COP versus efficiency section of the primer.

D. Water and fuel appendix (claim 5). - Puharich ("Cutting the Gordian Knot" and US 4,394,230): 0.9% saline at 600 Hz; 100 mW (V and I as separate rms readings, "A wattmeter was not used"); measured 16.20 cc of gas in 30 min (gas chromatography and mass spectrometry, Gollob Analytical, ±10%); "credit" 206.85 J (ΔH basis) against "debit" 180 J, a claimed surplus of 26.85 J, about 15 percent [read: digests 80-82, pp.1902-1931]. QE check [calc, .venv]: 10.80 cc of H2 is 0.482 mmol of water split. At ΔG = 237 kJ/mol that needs 114 J; burnt back, it returns ΔH = 138 J, which is less than the 180 J debit. So on Puharich's own measured gas there is a net loss, not a surplus. Two slips produce the printed surplus. First, the "ideal" yield treats 17.01 cc of water vapour as 17.01 cc of mixed gas (11.33 cc H2), when it is really 17.01 cc of H2 plus 8.5 cc of O2; so the true Gibbs efficiency is 63 percent, not 91 percent. Second, the "credit" multiplies the 16.20 cc of mixed gas by 12.77 J per cc of water vapour instead of 8.50 J per cc of mixed gas. This needs confirming against the page images (equations not in the text layer). It is a good teaching case: a source that measured its gas carefully (two laboratories) and still drew a surplus out of bookkeeping, inside the ΔH/ΔG window the primer explains. Puharich is also said to have driven a motor home "hundreds of thousands of miles" on water (p.1902; editorial, no source). - Hasebe US 4,105,528: data table of 2.8 V, 30 A and 116 cc/s of gas, "hydrogen 2.6 cc per coulomb", "over 20 times" Faraday [read: digest 82 p.1936]. QE check: Faraday gives 0.116 cc of H2 per coulomb at STP [calc: 22,414/(2 × 96,485)], so 2.6 cc/C is 22 times. 116 cc/s at 84 W is 0.72 J per cc of mixed gas, about one tenth of the Gibbs floor of 7.06 J/cc (7.06 kJ/L) [calc]. This falls under the arbiter's existing claim 5 reasoning but is a separate, checkable, patentee-stated measurement. - Chambers (Xogen) US 6,126,794: 12 V at 300 mA (3.6 W) pulsed, and a one-cylinder engine "successfully run ... without any gasoline" [read: digest 82 pp.1938-1952]. A 3.6 W input cannot supply an engine's fuel energy [calc: of order 1 kW of fuel for a small engine; memory, medium]. - Spiros WO 9528510, "the Availability of Over-Unity Energy": the claimed loop is high-pressure self-compression (to "150,000 atmospheres") plus a fuel cell; 1.49 kW of H2 from 1 kW in at 2,000 °C (using Bockris's calculated cell voltages, not measurements); the inventor's own tests ran to 2,450 psi [read: digests 84-86, pp.2003-2037]. Spiros defines "over-unity type (ii)" as 10 W in and 8 W of gas plus pressure. Never reviewed. - Garrett (1935), Blue, Aguero (65 A at 87 V pulses; chlorine and NaOH by-products from brine), Horvath (V8 on electrolysis gas), Eccles (field cell with electrodes outside the water, 25 kV) [read: digests 82-84]. - High-mpg carburettors: Pogue (three patents), Newberry, Shelton ("approximately eight times the mileage"), Schwartz ("by over 200%"), Tucker, Ogle ("in excess of one hundred miles per gallon"), with Kelly's repeated note that "the oil industry does not want functional high mpg carburettors" [read: digests 86-88, pp.2041-2087]. Booster articles: Hotsabi 18 to 27 mpg; a Saturn SL at 26, then 44-46, then 88 mpg "on a careful run"; Les Pearson's account that ECUs adapt back (pp.2280-2313) [read: digests 95-96]. An energy-bound check is possible: the heating value of the fuel against the road load at a stated speed. None was done. - Internal conflict: Papp's demonstration is "300 HP for 40 min" in the p.37 table and 35 minutes on p.642 [read: extracted text line 1653; physicist citing p.642]. Add it to the list of conflicting numbers.

E. Medical and other non-energy claims (Chapters 11-12, 14, 20-21). Absent from claims.md, the arbiter and the primer. - Beck: blood electrification described as a cure for HIV, cancer and other diseases (pp.999-1004, deep-read). Kelly then builds Beck's pulser as the design project in the Chapter 12 tutorial, with a treatment protocol (two hours a day for about two months; "50 to 100 µA through blood is essential") [read: digests 43, 45-46, pp.1060-1071]. - Chapter 20: autism "can be quickly combatted" with suramin (p.1375). A home-built Lakhovsky oscillator (12 kV neon transformer, spark gap, 500 kHz-1 MHz) with suggested settings "based on very limited patient experience" for "brain tumour" and "liver or pancreas tumour" (p.1413). An audio track said to reverse Alzheimer's (p.1416). Colloidal silver said to "cure both cancer and malaria" (p.1417) and "AIDS and Ebola" (p.1212) [read: digests 52, 58-59]. - Chapter 21: Keshe "GaNS" water said to reverse genetic modification in seed (pp.1418-1423, deep-read). - Kelly's essays on finance (mortgages invalid), the media and climate (pp.993-1005, deep-read). - These are outside the physics review. Under SOURCE HONESTY they are pertinent material in the source, and the reader should know they are there. The arbiter and primer should at least record them in one paragraph, say they were not reviewed (no medical reviewer), and keep the book's own safety warnings (pacemakers within 10 m, lethal capacitor charge; p.1408). The present silence could read as an endorsement of the rest of the book by omission, or hide the fact that the book carries such claims.

F. Sweet record details (appendix, deep-read but dropped later). - Maurice's statements that Sweet held an MIT MSEE (1969, thesis "Dynamics of Magnetic Domains") are recorded in the deep read (line 73) but not in the reviews. They are a cheap Investigator check (MIT DSpace thesis index). - Further statements by Maurice: a VTA said to have gone to "Senator John Glen", then to the Department of Energy, then to General Motors; a planned $200 million suit; people "claiming to be from the FBI" removing equipment on the morning after Sweet's death [read: digest 94 p.2243]. The contrarian mentions these. The arbiter keeps only "two unknown visitors". All of these are single-source statements by an interested associate. They should be kept beside the death account as that associate's statements. The National Archives, DOE and FBI FOIA records are the deciding records.

4. Accuracy spot checks (against _text/PJKbook-9cc3a5.txt)

QE also rechecked these arbiter/primer numbers: 5,740 C/L, the Gibbs floor 1.96 Wh/L and the thermoneutral value 2.36 Wh/L; the 0.0627 V/cell answer to Check-yourself question 1; the dipole effective area 1.2 × 10⁴ m² at 1 MHz and 0.3 W at 0.1 V/m; the Benitez stored energy 0.16 J and 160 W. All are correct [calc].

5. Process

• Source tags. These are present and honest throughout the reviews, arbiter and primer. [calc] items are reproducible, and [memory] items carry a confidence. The deep reads tag by page.
• Web checks. The physicist used 2 of about 6 web checks and left one [memory, medium; web check not done] (HHO boosters, physicist.md line 107). The arbiter's web check 4 covered it. The mathematician used 2 of 6 and the contrarian 8; the arbiter used 8 of 8. Taken together this is acceptable. The physicist's shortfall is a minor process defect that was compensated downstream.
• Verdict scale. It is applied consistently, and split verdicts are clearly separated. "TC 2800 list provenance unconfirmed" in the arbiter's brief table is a status, not a scale verdict. That is acceptable because the scale verdict ("Never properly tested") is given in the body.
• Primer against STEP 4 and the reader profile. All nine sections are present in order. Symbols are defined. The worked example is excellent and reproduced by example.py. Two gaps: (BH)max is used without saying what B and H are (one sentence would fix it), and the primer is silent on sections 3A (magnet-only motors) and 3E (medical claims).
• example.py runs cleanly (exit 0, .venv/bin/python). Its output matches primer section 5 (0.1598 J, 159.8 W, 52.9 A peak at 2.43 µs, 0.282 A average, Q = 0.667, 47.4 kHz).
• Corrections process. The arbiter's and primer's "Corrections to earlier documents" sections are well formed. Check before the recheck whether the orchestrator has applied them to claims.md and facts.json (claims.md still shows the old claim 4 and claim 7 wording and the "rectifier" wording for Moray, which the arbiter corrected).

6. Retention check

The question here is whether pertinent material from the sources or the Investigator was left out of the arbiter or primer because a reviewer judged it negatively.

• No case of material dropped because of a negative judgement was found. The arbiter keeps unfavourable-to-mainstream material carefully: Eyring's "found no fault", the Murdoch thesis's 18 percent case, the Michaels letter, the TC 2800 image, both readings of Meyer's death, and both Papp casualty accounts.
• Dropped without reason (minor defects): 1. Lyden US 11,588,421 B1 (granted 2023), an Investigator lead. It cites Benitez's GB patents as keeping two batteries charged "with relatively little or possibly no energy loss" and cites Moray's US 2,460,707. The physicist and mathematician recorded it as the applicant's characterisation. It is absent from the arbiter and the primer. It is a recent granted patent that relies on the very Benitez example the primer refutes, so it belongs in the record next to that example, with its stake stated. 2. Sweet's 1986 test outputs (Generation 1, 6-19 W; revision two, 100-110 W; first test 19 May 1986 shown to Goldes and Roberts; hyiq.org via sites.md). The mathematician and contrarian kept them; the arbiter kept only "Roberts's lab notes". These are the earliest and lowest numbers in the Sweet record and should sit beside the 1987 figures. 3. Goldes's stake in the primer. The arbiter names Mark Goldes (Aesop Institute) as funder with a stake, but the primer's evidence-table row for Sweet omits him. 4. Kelly's own caveats on Perendev, Tewari and the mains-rod circuits, and his true-power explanation (pp.1089-1091) (sections 3A-3C). These were never reviewed, so they were not dropped by judgement. They are Kelly's own corrective statements, and the reader profile benefits from them.
• Sections 3A-3F above are material that never reached the reviewers (sampling, not judgement). They must be added under SOURCE HONESTY, whatever verdict the arbiter then gives.

7. Remediation actions

Deep reads (most important first; at most 8)

1. PJKbook-9cc3a5, pp.2196-2238 (Opus). The Papp patent US 4,428,193 and Britt US 3,977,191. What exactly is the claimed mechanism (gas mix, thorium and other radioactive electrode contents, "minute fusion", 1,000-hour charge, coil and capacitor values)? What input power does the patent state (24 V battery, 27.12 MHz at 3.4-8.4 A)? Is any measured output given? Render the circuit pages if the values are garbled. Purpose: give the arbiter's Papp verdict the patent's own mechanism (fusion), not only recombination.
2. PJKbook-9cc3a5, pp.1901-1960 (Opus). Puharich (article and US 4,394,230), Hasebe US 4,105,528, Chambers US 6,126,794. Extract every input and output figure with its measurement method. Render the equation pages (pp.1909-1912, 1929-1930) and the Hasebe table (p.1936). Confirm QE's recomputation of Puharich's balance (true Gibbs efficiency 63 percent; ΔH credit 138 J against a 180 J debit; the 12.77 J per cc of vapour vs 8.50 J per cc of mixed gas slip) and Hasebe's cc per coulomb against Faraday. Purpose: claim 5 sub-verdicts, and a candidate second worked example (a surplus produced by bookkeeping).
3. PJKbook-9cc3a5, pp.42-146 (Sonnet). Chapters 1-2. List every quantitative claim with its source and any test: Wang and the "six-month Reliability and Safety testing", Johnson US 4,151,431, Perendev, Hohl/Ascanius, Adams (the COP 1,223 and 800% passages and who measured them), Kawai (already deep-read in the appendix), Tewari, RotoVerter, coil shorting, Hatem, UFOpolitics. Purpose: a new claim 3 sub-verdict on magnet-only motors, and correction of the facts.json Adams entry.
4. PJKbook-9cc3a5, pp.147-211 (Sonnet). Chapter 3a. Tseung frame (the COP 1.5 anecdote), Heins (patent COP 39.6 and 3.77; render pp.150-151), Markov EP 0844626, the Lenz-free transformer and the Noskills replication, Klingelhofer COP 8, Figuera (the 1902 Daily Mail), Alexkor, Barbosa and Leal (169 kW, COP 100; Clarence 3 kW), Matchett/Ascanius measurements, and Kelly's doubts. Purpose: claim 3 and claim 6 sub-verdicts (the earth-electron trap).
5. PJKbook-9cc3a5, pp.2003-2037 (Sonnet). Spiros WO 9528510. State the "over-unity" argument and render the data tables on pp.2018-2022 (OCR garbled). Which figures are measured (up to 2,450 psi) and which are calculated (Bockris, up to 2,000 °C and 150,000 atm)? Purpose: claim 5.
6. PJKbook-9cc3a5, pp.1375-1417 plus pp.1060-1071, 1203, 1210-1212 (Sonnet). List every medical claim (Beck protocol, suramin and autism, Lakhovsky settings for tumours, Alzheimer's audio, colloidal silver for cancer, malaria, AIDS and Ebola) with its page and stated source, and the book's own safety warnings. Purpose: a short, factual "medical claims in the book (not reviewed)" paragraph for the arbiter and primer.
7. PJKbook-9cc3a5, pp.1151-1277 (Sonnet). Chapter 14. Kelly's LM2587 converter "COP 1.5" (p.1259; how it was measured, meter type, steady-state or not), the Perkins and Frenette heaters ("at least 100% efficient"), Kim's 30 L/min HHO heater (4.2 kW by Kelly's own Faraday figure), the ram pump, Bhargava. Purpose: claim 2 (Kelly's own measurement) and the primer's COP versus efficiency teaching.
8. PJKbook-9cc3a5, pp.2041-2087 and 2280-2313 (Sonnet). The high-mpg carburettor patents (Pogue, Newberry, Shelton, Schwartz, Tucker, Ogle) and the booster mpg reports (Hotsabi, the Saturn SL). Extract each mpg claim, vehicle, speed and test method. Purpose: an energy-bound check against the road load and the heating value of gasoline (the arbiter has the HHO booster sub-verdict but no carburettor verdict).

Changes to later files (MINOR route)

• facts.json
• Correct the #ch1-2 Adams entry (27.6 W in, 33.78 kW out, COP 1,223, by forum builder "Maimariati"; p.93; also p.29, 27 W and 32 kW).
• Add Heins COP 39.6 (p.150); Klingelhofer COP 8; Barbosa and Leal 169 kW / COP 100; Wang's "six-month testing"; Kelly's LM2587 COP 1.5 (p.1259); Kelly's true-power passage (pp.1089-1091); Puharich, Hasebe, Chambers, Spiros and the carburettor figures from the deep reads; the medical claims; the Papp 35 min vs 40 min conflict (pp.37, 642).
• Change the how_read fields after the deep reads.
• claims.md
• Claim 3: add the magnet-only motor class.
• Claim 5: add Puharich, Hasebe, Spiros and the carburettors.
• Claim 6: add Barbosa and Leal.
• Add a one-line note that the book also makes medical claims (Chapters 11-12, 20-21) that are outside this review.
• Apply the arbiter's pending corrections (claims 4, 6 and 7) if the orchestrator has not yet done so.
• arbiter.md: add a "QE revision" section with: 1. a claim 3 sub-verdict for magnet-only motors (closed cycle: Refuted by calculation; the Wang test claim and Perendev output: Never properly tested; Kelly's own Perendev caveat kept); 2. claim 2 additions (Adams COP 1,223, RotoVerter, Heins 39.6, Klingelhofer, the South African COP 45 builder, Kelly's LM2587 "COP 1.5"; most likely Never properly tested, with Kelly's own measurement assessed against the switching-converter input-current problem); 3. claim 5 additions (Puharich's claimed 15 percent surplus, which on his own measured gas is a net loss (138 J back for 180 J in), Hasebe 22 times Faraday against the Gibbs floor, the Chambers engine on 3.6 W, Spiros, the carburettors); 4. Papp mechanism per the patent (fusion claim) under claim 6; 5. Barbosa and Leal under claim 6; 6. Lyden US 11,588,421 and Sweet's 1986 outputs restored; 7. a paragraph listing the medical claims as not reviewed; 8. the Papp 35/40-minute conflict.
• primer.md
• Section 6 evidence table: rows for magnet-only motors, Puharich/Hasebe and Kelly's LM2587 measurement.
• Section 7: quote Kelly's pp.1089-1091 true-power warning beside the "peaks and products are not power" lesson.
• Section 2: one sentence defining B, H and (BH)max.
• Add Goldes (stake) to the Sweet row.
• Add a short boxed note that the book contains medical claims that this primer does not assess.
• Optionally add the Puharich balance as a second short worked example (how a careful gas measurement still yields a bookkeeping surplus).
• Investigator (next run) leads:
• the 2008 Chinese "Reliability and Safety" test of Wang's generator;
• Perendev / Mike Brady's later record [memory, low: reports of legal action against Brady; to be checked];
• Sweet's MIT MSEE 1969 (MIT DSpace thesis index);
• the 1902 Daily Mail report on Figuera;
• Gollob Analytical Services and the Puharich measurements;
• FOIA records on the claimed Glenn / DOE / GM and FBI episodes in Maurice's account;
• a comparison of V34.8 with V32.9 for changed figures.

Corrections to earlier documents

(For the orchestrator, under CORRECTIONS ARE WELCOME. QE edited no other file.)

1. facts.json PJKbook.pdf#ch1-2-pp42-146, summary. "digest records one replication with 27 W electrical output" is wrong. The text says the builder's "input power is 27.6 watts and the output power is 33.78 kilowatts", a claimed COP of 1,223 [read: extracted text p.93]. The digest (digest 04, p.93) has it right. The error came from the fact-finder's sampling.
2. facts.json #ch3a (Heins). Incomplete rather than wrong: the patent test figure "0.29 watts ... 11.51 watts, which is a COP of 39.6" on p.150 is missing [read: extracted text p.150].
3. Arbiter, list of internal number conflicts (via claims.md cross-cutting facts). Add Papp "300 HP for 40 min" (p.37 table) against 35 minutes (p.642) [read: extracted text line 1653].
4. Book misprint worth a footnote where the lux test is cited. On p.476 the row "180 mA 200 lux 0.936 watts" gives "101 lux per watt"; it should be 214 [calc]. The arbiter's range 139-221 is unaffected.

Recheck

Recheck run 2026-10-09 against the current claims.md, facts.json, arbiter.md, primer.md and example.py. Checked by search and by reading the changed passages; example.py was re-run.

Other checks. example.py exits 0. QE's own figures that the deep reads changed (Puharich wattmeter wording, the 63 percent assumption of STP, Hasebe floor 7.05 J/cc, Spiros litres per hour) are corrected in claims.md and the arbiter, so no stale QE number is relied on. The arbiter states that where the QE revision and the body differ, the revision rules.

Items 16-18 are optional or by design and do not affect any verdict.

Final: PASS