Naudin-JL: Quality engineering (Step 4b)

QE run 2026-10-09. Inputs checked: _text/manifest.json (55 documents, 62 chunks), all 62 digests (read in full), the 12 deep reads in deep/, facts.json, claims.md, sites.md and research/ notes and pages, physicist.md, mathematician.md, contrarian.md, arbiter.md, primer.md and example.py. Tags: [read] = checked against the extracted text or a deep read; [calc] = my arithmetic; [memory] = recalled, with confidence.

Verdict: MINOR

The central verdicts hold up. Every load-bearing number I checked against the text is right, and the arbiter's recalculations reproduce. The defects can be fixed by updating facts.json, claims.md, arbiter.md and primer.md: (1) several pieces of pertinent material that the reviewers had were dropped between the reviews and the arbiter or primer, mostly material that cuts for the inventor (the Cornille-Naudin Trouton-Noble anomaly, Honda's oil-immersed weight-loss test, Brown's air-versus-oil result, the CFR replications Naudin lists including the CNAM Paris one); (2) one shortcut: the Dstl slide charts were never read as page images; (3) the arbiter calls Naudin an "unpaid hobbyist", and nothing in the record supports that; (4) the primer leaves COP and about ten acronyms undefined. One lead could move the claim-4 verdict: the CNAM (Paris) replication presented at ICCF-12. If it reported excess heat, claim 4 becomes a split verdict. A QE revision to the arbiter can handle that, so Steps 2 to 4 do not need to be redone.

1. Coverage

• Manifest: 55 documents, 62 chunks; 62 digest files, one per chunk, no missing and no extra files [checked by script]. No document was skipped by Step 0. OCR pages are noted per document. The README lists the files the compiler did not fetch (images, videos, sub-pages); those are outside the folder, not skipped files.
• facts.json: 55 entries, one per manifest document; each how_read matches what was done (digest only, or digest plus a deep read with pages). The browser junk was excluded at retrieval, not in the manifest.
• Secondary sources are marked as such (Related-Research/, "[secondary]" in relevance). No earlier AI-written analysis is in the folder.
• Coverage: no defect.

2. Shortcuts

Deep reads exist for both patents and for 10 of the 17 secondary documents (ARL-TR-3005, NASA/CR-2004-213312, Tajmar, Musha, Horky, Jelinek-Pekarek, Martins-Pinheiro, Ianconescu, Losonc, Bearden's MEG paper). All of Naudin's own HTML pages are digest-only. They are short single pages, and the reviewers and arbiter quote them from _text/ directly. For the central numbers I checked that text myself (section 4). The MEG screen values (6.764 W, 10.28 W, 255.3 V x 27.23 mA) trace to the Horky deep read's page images of Naudin's screens. Acceptable.

Defects: - S1. Dstl slides (claim 1a), charts not read. The only institutional dissent kept against the ion-drift account is Dstl slide 37 ("calculated ionic thrust more than an order of magnitude less than that observed"; "weight loss observed before lifting"). The arbiter and primer keep it as "method not shown". Slides 17 and 28-34 hold the Dstl lifter dimensions and the voltage, current, power and mass data that would let a reader test that statement against F = Id/μ. They are charts that were never rendered (facts.json: text_available "partial"). This is a central-claim item resting on the digest alone. - S2. Horky pp.1-4 not deep-read. The deep read covered pp.5-29. Pages 1-4 contain two statements relevant to SOURCE HONESTY that no later file weighs: "Naudin is now a member of Bearden's team" and Bearden's reported 2.5 kW closed-loop automotive MEG modules (four combined into 10 kW, through a "National Materials Science Laboratory", probably in Hungary) [read: digest Horky p.2]. Both bear on Naudin's stake and on the MEG absence-of-evidence analysis.

3. Missed material (from my own reading of the digests and research files)

Items present in the sources or the reviews but missing from the arbiter or primer (several are also retention defects; see section 5):

1. CFR replications Naudin lists [read: digest 2003-CFR-plasma-electrolysis-reactor-index p.1]: "Full CFR replication at the CNAM in Paris" (19 Sep 2006), said to be presented at ICCF-12 (Yokohama, Nov-Dec 2005); Ronald Hohls (South Africa, Mar 2006); Boulder (Colorado); "a French lab"; Morgan H. and Marissa C.; a CFR-powered Stirling engine. Physicist and mathematician mention CNAM, South Africa and Boulder; the arbiter and primer do not.
2. Naudin's CNAM collaborators [read: Kowalski item 73 p.1]: Naudin thanks Prof. Pierre Clauzon, Prof. Jean-François Fauvarque and Gérard Lallevé of the CNAM Laboratoire d'Electrochimie Industrielle "for scientific support and help with CFR test runs", and tells Kowalski that "some French physicists work closely with me". No review file names them. Lead: I recall a Fauvarque, Clauzon and Lallevé paper on the thermal behaviour of plasma electrolysis at ICCF-12 [memory, medium; not checked], probably on lenr-canr.org. This is the institutional replication behind item 1. Whatever it reports, it is the most neutral CFR test connected to Naudin and should be retrieved and weighed.
3. Cornille-Naudin Trouton-Noble experiment, 1996-99 (Investigator lead; physicist and contrarian discuss it): a charged 500 pF capacitor on a nylon thread, about 70 kV, reported to "lock" East-West over dozens of repeats; this is Naudin's own experimental work and the basis of Cornille's "stimulated forces" theory [read: research/2026-10-09-google-search.md]. Dropped from the arbiter and primer.
4. Brown's letter (ARL Appendix B), captured in deep/ARL p.36: a 4 ft rotor gave "virtually the same" torque in air and under transformer oil up to about 300 kV, "seeming to prove that aero-ionization was not wholly responsible"; years of torque logs that Brown says correlate with solar and lunar time and sidereal time; the Navy declined a proposal on the expectation of no vacuum effect; the 15-inch dome that lifted 110% of its weight [read: digest ARL p.36-37]. The arbiter keeps only the vacuum part.
5. Honda R&D 1996 (via Musha): a 62 g glass-disc capacitor immersed in oil to suppress corona, 8 kV AC pulses, weight loss 0.9 ± 0.63 g [read: deep/Musha]. Physicist and mathematician discuss it; the arbiter and primer drop it.
6. NASA/ISR low-pressure details [read: deep/NASA-CR p.11-13]: one event at 5.5e-5 Torr, 44 kV, Device 2 moved about one-eighth of a turn together with a large arc (the authors attribute it to ejected material); motion of less than a tenth of a turn at intermediate voltages with "the air ... becoming charged"; gaps were not designed for low pressure, and the apparatus could not resolve less than one-sixth of a turn; thrust in nitrogen and argon at moderately reduced pressure. The primer's "thrust vanishing below about 300 Torr" overstates a coarse null and omits the one event.
7. Transdimensional Technologies vacuum test: Naudin's ACT page says TdT tested the NASA ACT "in June 2001 ... in the vacuum chamber of NASA MSFC" (photo, no result) [read: digest 2001-ACT-NASA p.1]. This is a third Huntsville vacuum test alongside Gravitec/NSSTC 2003 and MSFC 2004, and it bears on arbiter open question 7.
8. Naudin's untested Lifter sub-pages that bear on claim 1a: "The TT.Brown EKP device in MINERAL OIL" and "Test of an Unlinked Lifter and fully enclosed in a box" [read: digest 2001-Lifter-performance-data-index p.1]. The source-gap note lists the box page but not the mineral-oil page. If Naudin's Brown device worked in oil, it bears directly on whether the effect needs air.
9. Recent work cited in the folder's Wikipedia copy, not weighed: Tajmar, Kößling and Neunzig, "In-depth experimental search for a coupling between gravity and electromagnetism with steady fields", Scientific Reports 14, 19427 (2024) [read: digest Wikipedia p.1]. It is relevant to claim 1c.
10. GEGENE closed-loop and calorimetric test titles (claim 7): TEST #3 "calorimetric method", TEST #4 "first test of a closed-loop GEGENE by Romero UK", TEST #9 "CLOSED-LOOP Phase 2 via Grid Tie Power Inverter" [read: digest GEGENE p.1]. The arbiter stresses that a closed loop is the cheap decisive test (for the MEG), so claimed closed-loop tests in Naudin's own record should be listed as claims not obtained, not left out.
11. Naudin's home page: "direct conversion of the environment temperature into electricity (COP up to 1.35)" (Zaev) and "The Cornille's electrostatic pendulum, now demystified" [read: digest nd-index-JLNLabs-home p.1]. These are minor claim-7 items.
12. Orbo page, both sides: Naudin writes "4.75 is NOT A COP" but on the same page says the loading test "demonstrated an overunity effect" [read: digest Orbo p.1]. The arbiter cites only the candid half. SOURCE HONESTY requires both.
13. 2SGen and DLE self-reports (claim 7): with the magnet, the DC input current drops; "best tuning ... no change in measured DC input power while the output coil is loaded"; DLE-TEST25 "input power drops dramatically and RPM doubled" under load. These are in facts.json; the arbiter's claim-7 list should name them.
14. EarthTech detail: the 0.98 average COP2 excludes runs 9-11 (stirrer off) [read: research/pages/et_replicationjln_2003.txt]. State it wherever 0.98 is quoted.

4. Accuracy spot checks

example.py runs without error in the project venv and prints every primer number in sections 5 and 4 (side check), including the strip-restricted pull of 0.91 to 69 mN [read: run output].

Errors found: - A1. physicist.md (claim 3 verdict) and arbiter.md (claim 3) say the MEG closed loop "has been missing for 24 years". From Naudin's 13 Nov 2000 note to Oct 2026 is about 26 years [calc]. - A2. primer.md s.3.1: "NASA/CR p.11 thrust vanishing below about 300 Torr" overstates the result (item 6 above). It should read: no rotation resolved below 300 Torr on an apparatus that could not resolve less than one-sixth of a turn, with one arc-linked eighth-turn event at 5.5e-5 Torr; devices not designed for low pressure. - A3. arbiter.md "Stance on evidence": Naudin as an "unpaid hobbyist" has no source. The record shows Naudin as co-inventor on two patents assigned to AES, a company "created 1998-99 to develop and sell electromagnetic generators" [read: research/2026-10-09-pappers-societe.md]. An unopened snippet calls him its R&D director, and Horky says he joined Bearden's team (S2). Naudin himself says he works "in my own and private lab" [read: Kowalski item 73]. contrarian.md repeats "unpaid hobbyist". The primer's wording (no sale found; a snippet names him AES R&D director) is the accurate one. - A4. primer.md s.9: Naudin's pages are given as "jlnlabs.online.fr". The main site is jnaudin.free.fr; jlnlabs.online.fr holds the CFR and MAHG pages [read: README p.1]. Give both.

5. Retention check

The reviewers did not suppress evidence: each discusses the material it judges weak and gives a reason. The loss happens downstream, in the arbiter and primer, and it falls mostly on material that favours the inventor. Each item below was in a review or a deep read, is pertinent, and is absent from both arbiter.md and primer.md:

• R1. Cornille-Naudin Trouton-Noble "East-West locking" anomaly and Cornille's "stimulated forces" theory (physicist claim 6d, contrarian claim 1d). This is Naudin's own experimental work and his co-inventor's theory for the patents and the Lifter. The physicist judged it a "reported anomaly, single-source ... I do not rule it out"; it then vanished downstream. Defect.
• R2. Honda R&D oil-immersed capacitor weight loss (physicist claim 1: "I do not consider it a test"; mathematician checked Musha's arithmetic). It was dropped because it was judged weak. It belongs in the arbiter's claim-1c record as a single relayed measurement with large scatter. Defect.
• R3. Brown's air-versus-transformer-oil torque result and his solar, lunar and sidereal correlations (deep/ARL). The arbiter lists "Brown's undated letter" only as a vacuum claim. Defect.
• R4. CFR replications Naudin lists (CNAM Paris 2006 at ICCF-12, Hohls, Boulder), which the physicist and mathematician mention as self-reported. The arbiter's claim-4 "Positive reports kept in the record" omits them. Defect.
• R5. NASA/ISR eighth-turn vacuum event and the low-pressure design caveat (deep/NASA-CR). The arbiter and primer cite ISR only as a null. Minor defect.
• R6. Orbo page wording "demonstrated an overunity effect" beside "4.75 is NOT A COP". This is adverse to the candour reading; SOURCE HONESTY cuts both ways. Minor defect.

Investigator leads (sites.md "Leads for review"): all others are retained in the arbiter (Porcelli, Buhler, Flickinger, Bearden's 7.3 and DoD replication, Colmac, AES, EarthTech, Kowalski-Little-Luce, Mizuno 2004, MSFC/ISR/Campbell, Tajmar-Schreiber, Bishop, Steorn CEO, BiTT/ReGenX). The exception is the Cornille 2020 review and Trouton-Noble lead (R1).

6. Process

• Source tags: present and honest throughout; [calc] is used for recomputation; the single-source items are described, not labelled. The arbiter and primer correctly report interested parties as their own characterisations, except A3.
• Web checks: physicist used 3 of about 6, mathematician 1, contrarian 1, arbiter 4 of 8. Each lists the [memory] items left unchecked and states that none drives a verdict. On inspection that is true: (BH)max, H₂ bond energy and switch-mode efficiency are high-confidence textbook values, and the medium-confidence items (thermocouple tolerance, fair-weather current, THS720P being a handheld) are side remarks. The failed checks (Stein, Tajmar-Schreiber full text) are reported as failed. Acceptable.
• Verdict scale: applied consistently. Split verdicts (1a to 1d, 3a to 3c, 5a and 5b, 6a and 6b) use the scale names exactly. The arbiter's reasoning for claim 4 ("Tested, not supported" rather than "Refuted by calculation", because the cooldown practice rests on a private communication) is careful and correct.
• Primer and STEP 4: all nine sections are present in order, plus a worked example with example.py, an evidence table, five questions with answers, and further reading. The source gap and interested-party notes are repeated. No new verdicts.
• Primer and reader profile, defect P1: COP is never defined (it first appears in s.4 as "apparent COP"). These acronyms are used undefined: ARL (Army Research Laboratory), Dstl (UK Defence Science and Technology Laboratory), ISR (Institute for Scientific Research, NASA's contractor), MSFC (Marshall Space Flight Center), NSSTC (National Space Science and Technology Center), AES (Advanced Electromagnetic Systems S.A.), CFR (Naudin's "Cold Fusion Reactor"), ICCF (International Conference on Cold Fusion), DoD, VTOL, TM (Technical Memorandum). MAHG is expanded only in the source-gap context.
• example.py: runs, and its output matches the primer.

7. Corrections to earlier documents

1. physicist.md (claim 3 verdict) and arbiter.md (claim 3, "apparent COP" paragraph): "24 years" should be "about 26 years" (13 Nov 2000 to 2026). Evidence: date arithmetic.
2. arbiter.md "Stance on evidence" and contrarian.md section 2: "unpaid hobbyist" is unsupported. Replace it with: "no sale of devices found; co-inventor on two patents assigned to AES, a company formed to develop and sell electromagnetic generators; one unopened snippet names him AES R&D director; Horky (2009) states, without source, that he joined Bearden's team; Naudin describes his lab as 'my own and private lab'". Evidence: research/2026-10-09-pappers-societe.md; Kowalski item 73; digest Horky p.2.
3. primer.md s.3.1: replace "thrust vanishing below about 300 Torr" as in A2. Evidence: deep/NASA-CR p.11-13.
4. primer.md s.9: give jnaudin.free.fr as Naudin's main site (A4). Evidence: README p.1.
5. claims.md source gap: add the "TT.Brown EKP device in MINERAL OIL" page to the Lifter sub-pages not obtained. Evidence: digest 2001-Lifter-performance-data-index p.1.
6. Wherever "EarthTech COP2 average 0.98" appears (arbiter claim 4, primer s.4 and s.6, physicist, mathematician): add "excluding runs 9-11, stirrer off". Evidence: research/pages/et_replicationjln_2003.txt.

8. Remediation

Deep reads (most important first; at most 8)

Investigator leads (records, not deep reads)

• The CNAM / ICCF-12 CFR paper (Fauvarque, Clauzon, Lallevé; lenr-canr.org, ICCF-12 proceedings) [memory, medium on the paper's existence].
• Naudin's "TT.Brown EKP device in MINERAL OIL" and "Unlinked Lifter fully enclosed in a box" pages (jnaudin.free.fr, Wayback).
• Tajmar, Kößling and Neunzig, Sci. Rep. 14, 19427 (2024); arXiv:2402.15640.
• GEGENE TEST #3, #4 and #9 pages; Cornille, "Replication of the Trouton-Noble Experiment" (Academia.edu, AES address).

What each later file must change

• facts.json: on the CFR index entry, add the replication list and CNAM collaborators (deep reads 3 and 4); on the Horky entry, add the p.2 statements on Naudin and Bearden's modules with their source status (deep read 2); on the Dstl entry, add the chart values and the result of the F = Id/μ check (deep read 1); on the Orbo entry, quote both of Naudin's statements; on the Lifter performance index entry, add the mineral-oil page as not obtained.
• claims.md: apply correction 5. In claim 4, name the CNAM collaborators. In claim 7, add the GEGENE closed-loop test titles and the Orbo "demonstrated an overunity effect" wording.
• arbiter.md: add a "QE revision" section that:
• replaces "unpaid hobbyist" (correction 2) and fixes "24 years" (correction 1);
• in the claim-1c vacuum record, adds the Honda/Musha measurement (single relayed source, 0.9 ± 0.63 g, oil-immersed, AC pulses, no controls described), Brown's air-versus-oil torque result and his time correlations, the ISR eighth-turn arc event with the design caveat, the June 2001 TdT/MSFC vacuum test (result unknown) and the 2024 Tajmar et al. study as a lead;
• adds the Cornille-Naudin Trouton-Noble anomaly and "stimulated forces" to claims 1 and 6, described as reported by the co-inventor with no raw data and the controls not described;
• adds the Naudin-listed CFR replications and the CNAM link to claim 4's positive reports. If deep reads 3 and 4 and the Investigator's retrieval show that CNAM measured excess heat with a droplet control, revise claim 4 to a split verdict; otherwise state that the CNAM result is not yet obtained;
• reports the Dstl check result (deep read 1) under claim 1a;
• quotes both Orbo statements and lists the GEGENE closed-loop claims under claim 7;
• adds the stirrer note to EarthTech's 0.98.
• primer.md: define COP at first use and expand every acronym listed in P1; apply corrections 3 and 4; in the s.6 table, add Honda/Musha and the TdT test to row 1c and the Naudin-listed replications (CNAM, Hohls, Boulder; self-reported, reports not obtained) to row 4; add the Trouton-Noble claim to s.7 "Genuinely open" with its caveats; carry any verdict change from the arbiter's QE revision.
• contrarian.md: correction 2 is recorded here for the orchestrator. Reviewer files are not rewritten under MINOR.
• sites.md / next Investigator run: the four leads above.

Recheck

Recheck 2026-10-09, STEP 4b, against the current arbiter.md (QE revision section), primer.md, claims.md and facts.json. Checked by grep and by reading the changed passages.

• S1 (Dstl charts not read): fixed. arbiter.md lines 176-180 report the deep read of pp.17 and 28-34 and the F = Id/mu check (ion bound 2 to 5 times the weight lifted; the "order of magnitude" statement would need d of about 1 mm); primer.md s.4 uses it.
• S2 (Horky pp.1-4): fixed. arbiter.md line 212 quotes the "member of Bearden's team" statement as unsourced; claims.md and primer.md carry it; the 2.5 kW / 10 kW modules are listed as unsourced.
• A1 ("24 years"): fixed in arbiter.md (now about 26 years, correction logged at line 224). physicist.md line 137 still says 24 years; this is a reviewer file, left unrewritten under MINOR and recorded in the arbiter's corrections list. Not counted as open.
• A2 (300 Torr overstatement): fixed. primer.md line 57 now states the coarse null, the one-sixth-turn resolution, the design caveat and the 5.5e-5 Torr eighth-turn arc event.
• A3 ("unpaid hobbyist"): fixed in arbiter.md and primer.md (AES co-inventor, snippet, Horky, "private lab"). contrarian.md line 27 still has the phrase; reviewer file, recorded in the arbiter's corrections list. Not counted as open.
• A4 (site names): fixed. primer.md s.9 gives jnaudin.free.fr and jlnlabs.online.fr.
• R1 (Trouton-Noble): fixed. New row 1e in arbiter.md and primer.md s.6, plus primer s.7 "Genuinely open".
• R2 (Honda/Musha): fixed. In arbiter.md and in primer.md row 1c.
• R3 (Brown oil result and time correlations): fixed. arbiter.md line 187 and primer.md row 1c.
• R4 (Naudin-listed CFR replications, CNAM): fixed. arbiter.md and primer.md row 4 list CNAM, Hohls, Boulder, the French lab, Morgan H. and Marissa C., and the Stirling engine as self-reported. The CNAM lead was followed: the Fauvarque, Clauzon and Lallevé 2005 paper (about 1.35 at 350 V, up to 1.41, after cathode preconditioning) is included as a positive report, and claim 4 was changed to a split verdict, as the QE section allowed.
• R5 (ISR eighth-turn event): fixed (see A2).
• R6 (Orbo both statements): fixed. arbiter.md line 44 and line 217; claims.md.
• P1 (undefined acronyms): fixed. primer.md has a "Terms and acronyms" box defining COP, ARL, Dstl, ISR, MSFC, NSSTC, AES, CFR, ICCF, DoD, VTOL and TM (checked by string search).
• Missed-material items 1-14: all present in the arbiter's QE revision or the primer (CNAM, TdT June 2001, mineral-oil page in the source gap, Tajmar 2024, GEGENE closed-loop titles, Orbo, EarthTech runs 9-11 exclusion, Dstl, Horky).
• Corrections 1-6 (section 7): applied to arbiter.md, primer.md and claims.md; items 1 and 2 on physicist.md and contrarian.md are logged for the orchestrator, as agreed.
• Consistency: the primer's claim-4 row and the arbiter's claim-4 verdict agree (split verdict). No new unsupported statement found in the changed passages.

Orchestrator addition (2026-10-09): primer section 1a added at owner request (survey of research projects). Eleven projects, one table row and note each; verdicts copied from arbiter.md (including its QE revision), no new research or verdicts.

Final: PASS