Naudin-JL: Science Contrarian review (Step 2)

Reviewer: Science Contrarian. Date: 2026-10-09. Standard: the same PhD-level rigour as the physicist and mathematician, judging by experimental evidence rather than by consensus or reputation, and applying the same rigour to the inventor and to his critics.

Tags: [read] from the project's documents (extracted text in _text/, deep reads in deep/, Investigator files in research/); [web] from the Internet with URL; [memory] recalled, with confidence; [calc] my own arithmetic from [read] numbers (run in the project venv).

0. Source gap and how this review was done

The Step 1 source gap stands. Naudin's pages were saved as text, so the scope screens, graphs and run tables are mostly images that were not retrieved. Where a verdict below rests on Naudin's description of a result and not on the result itself, I say so.

I read: facts.json, claims.md, all 12 deep reads, sites.md and the Investigator's notes, and the primary text of Naudin's MEG v2.0 and v2.1 pages, the two Lifter vacuum-tube pages, Lifter1, Lifter v4, the Lifter story, the Gravitec vacuum page, the MAHG index and one-hour page, and Smith's critique. From the Investigator's saved pages I read the CFR calibration, low/medium-power and results pages, the MAHG test pages, and the full EarthTech replication of Naudin's reactor. I did not read the Bishop PDF itself (I rely on the Investigator's account and Smith's reply), the Dstl slides beyond slides 34-38, or Kowalski's ICCF-12 paper (Investigator's summary only).

Web checks (budget about 6, 1 used): the Stein (Purdue 2000) vacuum paper that Naudin and Losonc cite could not be found. The search returned only citations, one in the arXiv version of ARL-TR-3005 giving "Stein, W. B., Electrokinetic Propulsion: The Ionic Wind Argument, Purdue University Energy Conversion Lab, 5 Sep 2000" [web: https://arxiv.org/pdf/physics/0211001, search snippet]. The check failed, so Stein's vacuum result (0.31 mN at 17 kV, as quoted by Losonc) stays a second-hand report. No claim below depends on a medium or low confidence [memory] fact except where marked.

1. Corrections to Step 1

1. MAHG "COP 21" version label (claims.md claim 5). claims.md lists "21 (v2.1, 1 h, 5% duty, 51 Hz)". Naudin's own page for that run is titled "MAHG v2.0 - One hour tests with a COP of 21" (created 28 Jun 2005, updated 30 Jun 2005) [read: _text/2005-MAHG-v2-0-one-hour-COP21]. His index page files it under version "2.1" [read: _text/2004-MAHG-...index]. The labelling is inconsistent on Naudin's side. Reviewers should cite the run page title: v2.0.
2. "10 MHz pulses" (facts.json, Kowalski item 289 entry; claims.md claim 5 inherits Kowalski's numbers). The "about 250 V at 10 MHz pulses" is Kowalski's reading of a Pure Energy Systems article [read]. Naudin's own MAHG pages give pulsed half-sine DC at 50 % duty from a transformer (runs 11-37) and 51 Hz / 100 Hz square pulses from a battery and MOSFET (June 2005). "TGP110 - 10 MHz pulse generator" is the instrument's rating, not the pulse frequency [read: Investigator's saved pultests.htm; 2005-MAHG-v2-0-one-hour-COP21]. A reader should not take "10 MHz" as the MAHG operating frequency. This also matters for the Frolov-Moller source, where I could not check the frequency.
3. Claim 3, Naudin's own heat evidence is older and fuller than claims.md says. claims.md cites only the 2008 e-mail ("load resistor ... only a bit warm"). Naudin's page of 6 Dec 2000 already says: "the Rload warms up quikly when the MEG is switched on"; "in most of cases the 'apparent' power measured seems greater than the heat dissipated by Joule's effect in the RLoad"; "most of the power is radiated in EM form" (2.8 mT at 16 kHz near the load, 1250 V/m at 50 cm); and that calorimetric tests "make me pessimistic about the calorimetric tests results"; "unfortunately measurement artifacts remain possible" [read: _text/2000-MEG-v2-1-replication-tests]. The two statements (2000: "warms up quickly"; 2008: "only a bit warm") differ; neither gives a temperature. facts.json and claims.md should carry the 2000 text, including the EM-radiation explanation, which was an untested rescue of the energy balance.
4. Claim 1, Naudin's vacuum-tube pages say more than claims.md reports. On the "true Lifter" page Naudin writes that the plate is "attracted electrostatically by the dielectric medium (the air) outside the vacuum tube polarised by the strong electric field", that "the surrounding medium is therefore required for the motion", and "a true Lifter fully enclosed in a box will not move" [read: _text/2003-Lifter-true-lifter-wire-in-vacuum-tube]. So Naudin's own conclusion from the tube tests is that the thrust needs surrounding air, which is not a vacuum-thrust claim. claims.md claim 1 reads as though the tube pages support a non-air force. They do not. Also, the "true Lifter" test shows rotation on a torsion hanger of a 910 g assembly, not lift. The 25 mW hover is the unlinked test (plate 200 x 20 mm, 12 micrometre foil) [read].
5. Claim 2, qualification of the Ianconescu result (claims.md: "matched four measured lift points within about 6.5 %"). True for force. The deep read shows the model's current is 22-40 % above the measured current, so the measured thrust per ampere is 17-47 % above the paper's own ion-drift prediction F = I d / mu, and K1 carries a unit slip (mA/kV^2 should read microampere/kV^2) [read: deep, Ianconescu]. The match is real but is a fit with free-ish parameters (Deutsch assumption, Kaptzov hypothesis), not a parameter-free prediction.
6. Claim 4, source gap now partly closed. claims.md says the CFR run data "appear only second-hand in Kowalski item 73". The Investigator recovered Naudin's calibration, low/medium-power and results pages (research/pages/) and the EarthTech replication. The run table is still an image. The calibration ratio is printed on Naudin's page as "56525 / 55700 = 0.985" with the numbers written input/output; output/input is 0.985 and input/output 1.015, so the page's arithmetic line is inverted [read, calc]. For the medium-power run the page prints "23714 / 35767 = 1.51"; the ratio is output/input (35767/23714). Both inversions are harmless to the numbers but show the page was not proof-read.
7. Confirmed (no change needed). The ARL ion-wind correction (copper-ion 0.002 g recomputes to 0.023 g; ARL's drift estimate is the collisional form), the Martins-Pinheiro "simulation only" note, the Horky correction (artifact vs error) and the FR patent 2.5 / "divided by 3" / 2.77 note in claims.md all agree with the deep reads. Horky's COP > 4.5 is from his text ("Z toho plyne COP>4,5") [read: Horky text line 671] and equals 6.76 / 1.5 = 4.5 [calc].

2. Method note: what is "tested", what is "dismissed", what is "interested"

• Interested parties are named per claim. Bearden and Magnetic Energy Ltd hold the MEG patent; Cornille is Naudin's co-inventor and AES was the applicant of both patents; Serrano and Gravitec Inc. sell asymmetrical-capacitor thrusters; Bedini, Steorn and Moller / Frolov sell or promote their devices. Naudin himself is an unpaid hobbyist who says he works in a private lab; I found no sign that he sold devices, but his site promotes the ideas and he was a director of R&D at AES per an unopened snippet [web: Investigator's pappers note, greatdreams.com snippet].
• Silence is not contradiction. The Talley reports, Brown's vacuum notes, the Gravitec video, Stein's paper and Naudin's CFR run table were not obtained. I treat what they are reported to say as evidence of unknown quality, not as nothing.

3. Claim 1: Lifter thrust is not simply ion wind; some force exists in vacuum

a. Is the mathematics internally consistent?

Naudin offers no equations for the "electrokinetic effect on the medium" or the "electrogravitic" part; his own frame is verbal ("asymmetrical flow of EM energy", polarised dielectric medium). So there is nothing to check for internal consistency except his statements, which pull in two directions: the tube pages say the medium (air) is required, while the Gravitec page claims a medium-independent "electrogravitics part". The related mathematics by others: Ianconescu et al. is internally consistent apart from a unit slip and an unexplained current mismatch. Losonc's boundary-element result (net electrostatic force zero in vacuum) is largely guaranteed by construction (a closed charge system's pairwise Coulomb forces sum to zero), as the deep read notes, and its comparison sentence ("F2 about 20 mN is an order of magnitude smaller than 3.65 mN") is inconsistent with its own numbers [read: deep, Losonc]. The argument "no electrostatic force, therefore electro-gravity" is elimination, not derivation. Musha's ZPF model gets 1.28 g from chosen inputs (R = 1 mm, N = 1e24, omega 6e13) that differ from his own earlier frequency by about two orders; the constant 3.23e-13 could not be reproduced [read: deep, Musha].

b. Consistency with established experiments

In air (what Naudin calls the "main thrust"). Ion drift with collisional momentum transfer is quantitatively sufficient for his own lifters. My check with the ion-mobility value Tajmar uses, mu = 2.15e-4 m^2/(V s) [read: deep, Tajmar], and the wire-to-foil gap of 30 mm [read: Lifter1 and v4 pages], with the drift-thrust ceiling F = I d / mu [calc]:

Lifter1 leaves a factor of 2.5-2.8 of margin. Lifter v4 sits at the ceiling: at hover it needs an effective ion path of about 3.4 cm instead of 3.0 cm, which is plausible (the foil skirt is 40 mm high and the ions travel to its edge). I therefore cannot claim v4 exceeds ion drift. Equally, nothing in these numbers needs a second mechanism. This agrees with ARL (16 g drift estimate against a lifter of about 5 g [read: deep, ARL]) and with the ISR report (ion drift "consistent with all of our observations", force predicted and the dependence on ground position and polarity reproduced [read: deep, NASA/ISR]). The Dstl slide states that calculated ionic thrust is "more than an order of magnitude less than that observed", and that weight loss precedes lift; the slide gives no calculation [read: Dstl slide 37]. It conflicts with the ARL, ISR and my own ceiling numbers; either Dstl used the ballistic (collisionless) model, which ARL also found too small, or something else is unstated. It stays in the record as an unexplained institutional dissent, and the Dstl lifter-by-gas data (lift in air and nitrogen, none in helium at 4.2 mA / 12.8 kV) fit a gas-dependent mechanism such as ion drift as well as any other [read: Dstl slides].

The vacuum-tube tests (the "no corona" claim). In the unlinked test the plate is 200 x 20 mm of 12 micrometre aluminium, mass about 0.13 g and weight about 1.3 mN [calc]. The electrode is a tungsten wire inside a glass bulb at 25 kV, and the plate is in air a few centimetres away. Electrostatic attraction of an induced-charge conductor toward a charged body follows ~ (1/2) eps0 E^2 per unit area: for a field of 2e5 to 5e5 V/m over the 40 cm^2 plate that is 0.7 to 4 mN at full coverage [calc, crude upper estimate]. That is the same order as the plate's weight, so a plain electrostatic pull on a 0.13 g foil can account for the "hover", with no ion wind and no new force. What the test excludes is corona from the wire. It does not exclude corona or glass-surface charging at the plate edges or the bulb glass, nor the pull of the charged apparatus toward grounded room objects (the effect Tajmar designed his sealed-box test to remove). The "true lifter" version (wire and plate rigidly linked, rotating on a torsion hanger) is the better control, but the page gives no force, no angle, no repeat, no control with voltage off or the tube removed, and Naudin's own interpretation is that the surrounding air is required [read]. 25 mW (25 kV at about 1 uA) is also consistent: the ion-drift ceiling at 1 uA and 3 cm is I d / mu = 0.14 mN, below the 1.3 mN weight, so ion drift would not lift it, which Naudin takes as proof of something else. But the same electrostatic pull estimate above is 5 to 30 times larger than the ion ceiling, so the simplest reading is that these tests show electrostatic attraction, not ion thrust, and that is what Naudin himself first suspected [read: the "true lifter" page: "to check if the thrust observed previously was simply an electrostatic attraction ... or a real electrodynamic thrust"].

In vacuum. The record is mixed and none of it is clean:

• Positive reports, all weak in documentation: (i) Brown's 1973 letter (via Naudin, in ARL-TR-3005 Appendix B): vacuum thrust at 1e-6 Torr in Paris 1955-56, Winston-Salem 1957-58, G.E. 1959, notes "never published" and a film; he is the inventor and an interested party, and the letter is undated and has a blanked addressee [read: deep, ARL]. (ii) Stein, Purdue 2000, 0.31 mN at 17 kV at 1.3e-3 Pa, known only through Losonc and Naudin [read]; not found on the web [web check failed]. (iii) The Gravitec / NSSTC video (3 Jul 2003, 1.72e-6 Torr, 45 kV): rotation on a torsion wire, weaker than in air, unpublished "for confidential reasons" until Naudin posted it in Dec 2007; the video was not obtained, so I cannot see the pressure trace, current, arcing or outgassing [read]. Gravitec's president Serrano is a patent holder and seller of thrusters, an interested party. (iv) Porcelli and dos Santos Filho 2016 measured an anomalous force on dielectric asymmetric capacitors and fit a Clausius-Mossotti dipole account [web: https://digital-library.theiet.org/doi/full/10.1049/iet-smt.2015.0250, abstract only, Investigator]; this is a dielectric-capacitor measurement, not a lifter in vacuum. (v) Buhler / Exodus Propulsion reports about 2,000 vacuum-chamber runs showing propellantless force, unpublished data, commercial stake [web: Investigator's google note, snippets only].
• Negative reports: Talley (Veritay for the Air Force, 1988 and 1991) found "no thrust in the case of a static DC potential" and an anomalous force only during breakdown with current flowing, reported second-hand by Tajmar and ARL [read: deep, Tajmar, ARL]; the reports themselves were not found [web: Investigator, DTIC negative]. MSFC TM-2004-213283: "No NACAP performance was observed in vacuum" [web: https://ntrs.nasa.gov/api/citations/20040171466/downloads/20040171466.pdf, Investigator]. ISR / MSFC AIAA 2004: "no force was measured in a vacuum"; but the apparatus could not resolve less than one-sixth of a turn, had no torsion balance, and the gaps were set for air, so it gives no upper limit [read: deep, NASA/ISR]. Tajmar's own sealed-box test in air found any linear thrust below 10 uN with a corona discharge present, and he cites Talley for vacuum; he ran no vacuum test [read: deep, Tajmar]. Tajmar and Schreiber 2020: no force to +/-0.3 mg at up to 10 kV with no ionisation [web: https://www.sciencedirect.com/science/article/abs/pii/S0304388620300619, abstract only, Investigator]. A builder in Naudin's own replications log (Xie Xuejun, 17 Jun 2006) reports the lifter "went down as air was pumped out" and at very low pressure only glowed and did not move [read: _text/2001-Lifter-worldwide-replications-logbook]; that is a small, uncontrolled data point against an air-independent force, from inside Naudin's own community, and it deserves to stay in the record.
• The Campbell / NASA patent US 6,775,123 says thrust "is substantially and significantly reduced in vacuum" and adds a gas supply to operate in space [web: https://ntrs.nasa.gov/api/citations/20040120932/downloads/20040120932.pdf, Investigator]. This is NASA's own patent position and is in tension with Naudin's reading of the Marshall-site Gravitec test as "propellantless".

Measurement confounders that the vacuum positives have not excluded (my list, [memory] high confidence on the physics, general): outgassing and residual-gas discharge at 1e-6 Torr, which Tajmar proposes for Brown's results [read: Tajmar p.9]; field emission and leakage currents on insulators at 45 kV; electrostatic attraction of the charged arm to chamber walls; charged-torsion-wire effects; and thermal or mechanical coupling to the chamber. The negatives have their own gap: insufficient sensitivity and gap geometry.

c. Tested? By whom? Replicated?

In air: tested by ARL (devices lifted at 30-37 kV, no force readings), ISR / MSFC (RPM with friction-balance inference), Ianconescu (four lift points), Dstl, and hundreds of builders; replicated widely. The question "ion wind or not" in air was tested best by ISR (ground position, polarity, gas species) and by Tajmar's sealed box (null). Naudin's own control (enclosed box) appears on a page not in the folder ("Test of an Unlinked Lifter and fully enclosed in a box") and his tube page says an enclosed lifter will not move. In vacuum: tested, with no neutral controlled test of adequate sensitivity at high voltage with outgassing control on record. The nulls are low-sensitivity or second-hand; the positives are interested, undocumented, or unavailable.

d. What would have to change

If a static-DC propellantless force existed in high vacuum, the net-force theorem for an isolated charged system (momentum conservation with the field momentum included) would need an external agent: a coupling of the electric field to a preferred frame (Cornille's "stimulated forces", with absolute motion with respect to vacuum), to gravity (Brown, Musha), or to something else. That is a major change to classical electrodynamics and momentum conservation. I did not review Cornille's theory in detail; the Trouton-Noble-type claim ("capacitors sought an East-West position and stayed locked", 500 pF Plexiglas capacitor on a nylon thread, Wimshurst machine at about 70 kV [web: https://www.longdom.org/open-access/a-review-on-electromagnetic-propulsion-by-stimulated-forces-53822.html, Investigator]) is a candidate for the classic artefacts of charged bodies near grounded surfaces and for the torque of leakage currents and air currents, none of which the review says were controlled. Cornille is an interested party (co-inventor, AES).

Verdict (claim 1, split)

• Air: "the main thrust is not simply ion wind / a separate electrokinetic effect": Tested, not supported. Ion drift with momentum transfer to air reproduces the force, polarity independence, ground dependence and gas dependence in several controlled-enough experiments, and Naudin's own ceiling data (table above) are consistent with it. The Dstl remark is an unexplained exception, kept in the record. The vacuum-tube tests do not isolate anything beyond electrostatic attraction and leave air in the loop.
• Vacuum: a residual static-DC force without ions: Never properly tested. ARL named this "the most pressing question" in 2003 and recommended a systematic vacuum / pressure study [read: deep, ARL]. What exists is a handful of low-sensitivity nulls, one undocumented Air Force contractor null (Talley), and positives that are interested, unpublished or unobtained. It was not dismissed without testing; it was tested badly in both directions.
• "Electrogravitic" interpretation of any residual force: Not testable as stated, until a residual force with clean controls exists.

4. Claim 2: Lifters are real, reproducible and scale

a. Mathematics

Naudin's performance numbers are simple and consistent: Lifter1 18 W (40 kV, 450 uA) to hover, 23.9 W with 1 g payload; v4 81.4 W hover and 132.9 W flight; 133 W is 44 kV x 3.02 mA = 132.9 W [calc]. No errors found. (He quotes the power "required to compensate the weight" and a higher power "for a stable flight", which is sensible.)

b. Consistency with experiment

Yes, in air. ARL built three lifters and a wire that lifted at 30-37 kV (1.5 mA, 500 uA, 1 mA) [read: deep, ARL]; Ianconescu et al. computed lift for one lifter to 6.5 %; Naudin's log lists 351 self-registered replications to Oct 2006 [read]. Thrust per watt: Lifter1 gives 1.25 mN/W and v4 about 3.9 mN/W [calc]. Ion and Hall thrusters need 20-70 W per mN, i.e. 14-50 mN/kW [read: deep, Tajmar, ref 23], so a lifter looks far better per watt, but it pushes on ambient air and works only in an atmosphere, so the comparison does not carry over to space.

c. Tested and replicated?

Replicated widely, mostly by hobbyists. The "351 replications" count is self-reported and unverified [read]. The "Lifter Maximus" (194 g lifting 60 g) and the 250 g "Maximus II" are from Naudin's pages; I did not see the data pages (Lifter v2/v3/Maximus not in the folder). Scaling beyond the hobby range is not documented in neutral sources. I rate the claim as true for small lifters and unverified for payload scaling.

d. What would have to change

Nothing in mainstream physics. A lifter in air is electrohydrodynamics (EHD) of a corona discharge.

Verdict (claim 2)

Established for small lifters in air (lifting is reproducible, with several independent builds including government labs). Payload scaling to 194-250 g rests on Naudin's pages: Plausible, untested by neutral parties. What is not established is any link from this fact to claim 1's vacuum force.

5. Claim 3: Naudin's MEG replication gives "apparent" COP above 1

a. Mathematics

The inventors' paper (Bearden et al., 2000, 69 pp.) has an internal inconsistency that the deep read documented: the two scope pictures (Fig. 18-19, p.67) give an output/input power ratio of 12.7 / 3.588 = 3.54 per output coil [calc], but the paper's own fitted COP curve (Fig. 21) gives about 2.2 at the 51 V input of Fig. 18 [read: deep]. Input "High 51 V" x "RMS 117.5 mA" = 6.0 W, not the displayed 3.588 W [calc]. And "M RMS" is the RMS of the instantaneous power waveform, which for a sine-driven resistive load is 1.22 times the mean power, so 12.7 W corresponds to about 10.4 W mean [calc, deep]. A Tektronix support reply forwarded in 2002 says the RMS of a math waveform "is bogus and has no value" and "THERE'S NO SUCH THING AS RMS WATTS" [read: _text/2002-MEG-THS720P...]. The paper also gives no turns, wire, core size, frequency, duty cycle or load [read: deep]. The projected COP of 30-40 (Fig. 22-23) has "no basis" stated.

The critics' mathematics: - Jelinek and Pekarek (Sisyfos, 2010): mean output power equals mean input power for a lossless anhysteretic core in periodic steady state, regardless of the magnet or load. The deep read finds this sound in outline, with a sign error in eq (5), missing cross terms in (10) and a linear-core assumption in the appendices. The weak step is the claim that currents and voltages "do not depend on the magnet", true only for a linear core; the MEG operates by saturating a biased core [read: deep, Jelinek]. The energy conclusion does not need that step. - Bishop (2002): the Investigator's account says he computed P_in = 28 V x 0.125 A / sqrt(2) = 2.47 W using an "RMS power" formula, and concluded output matches input via the turns ratio [read: sites.md; Smith]. I side with Smith on the formula: RMS of a power waveform is not mean power, and Bishop's eq. (3) is wrong as Smith writes; Smith is also right that the input current sat on a DC level, which Bishop ignored, and that a "conditioned" 100 kohm carbon load is no longer 100 kohm [read: Smith]. But the same principle Smith uses against Bishop condemns the "M RMS" readings in Bearden's p.67 figures, which Smith's reply does not mention; Smith says Naudin used the mean of the power waveform, which I cannot check without Naudin's scope images [read]. Smith's own supplement, that precession energy in a 40 x 25 x 38 mm NdFeB magnet "of the same order as the field energy ... about 30 joules", of which 0.01 % per cycle at 25 kHz would give 75 W, is his speculation and is not supported by any measurement; his own number also shows the reservoir is finite (30 J at 75 W would last 0.4 s) [calc, read]. - Horky (2009): the deep read recomputes his published COP (1.57-2.89) as 1.37-2.53 on full supply power with doubling accepted and 0.69-1.26 for the one lamp actually measured; power factor never measured; the "input" is supply voltage times (I1 minus an undefined no-load current) [read: deep, Horky].

b. Consistency with experiment

A closed-loop self-running MEG would contradict the large body of transformer, hysteresis and energy-conservation experience, unless the magnet or the vacuum is a source. A finite transient COP above 1 from stored magnetic energy is not excluded by conservation, but any sustained gain is. Experimental facts in the record: - Naudin, 7 Nov 2000: "measured COP is 1.75" (MEG v2.0). v2.1: 6.76 W output against 3.25 W total DC input (COP 2.08), or against 1.5 W after subtracting the 1.75 W control-board draw (COP 4.5) [read, calc]. The no-load subtraction treats the control board as free; whether it should be subtracted is the same ambiguity as Horky's undefined I0. - Naudin could not close the loop (13 Nov 2000, and still "not yet" in 2008), the load resistor was cooler than the power implied, calorimetry was not done, his own coils arced and broke on 16 Nov 2000 at >3 kV [read]. - Naudin's own v3.0 note: the effect needs a "conditioned" 100 kohm load or a varistor, a very specific non-linear load. Flickinger (a MEG builder) says COP above 1 appears only with non-linear loads and that linear loads do not give Bearden's waveforms [read: Investigator, megnot01]. A result that exists only when the load is a non-linear resistor and only on a digital scope's math function is the textbook signature of a power-measurement error with non-sinusoidal waveforms. I say "signature", not proof: nobody has shown the artifact for Naudin's actual traces, and Naudin states the phase between voltage and current was checked with a second (analog) scope [read: v2.1 page]. - The Investigator found no sign of the university / DoD replication and the "government-certified" laboratory test that Bearden announced on 2 Dec 2000 [web: https://zpenergy.com/modules.php?name=News&file=print&sid=3918, Investigator]; Bearden said Naudin reached COP 7.3 in a third build. Naudin's folder pages show 1.75 and 2.08 / 4.5; the 7.3 is Bearden's number, an interested party's report, and I cannot confirm it from Naudin's pages. The patent was kept in force to term and pledged to Colmac Holdings (Canada) in 2006 [web: Espacenet, Investigator]; that is a record of continued ownership interest, not of a working product.

c. Tested? By whom? Replicated?

Open-loop signal replication by Naudin (and later Horky, who copied Naudin's v3.0). No calorimetry, no closed loop, no independent laboratory. Bearden's own data are the inventors' only. So: tested by the inventors and one hobbyist; not by a neutral party. The negative arguments (Bishop, Jelinek-Pekarek) are calculations, not experiments, and Jelinek-Pekarek did not re-analyse Horky's numbers [read: deep, Jelinek]. Absence of evidence: the announced independent tests vanish from the record; competing explanations are (1) never carried out, (2) carried out and negative, (3) carried out and confidential. The record cannot choose among them. The decisive records are DTIC / university reports naming the MEG, Bearden's cheniere.org Correspondence pages after 2001 and any Wayback copy of Naudin's v3.x pages (criticofsb.htm, megv3.htm) [web: Investigator].

d. What would have to change

A device that returns net electrical energy from a permanent magnet and a switched core would require either a finite, depletable energy source in the magnet (not an energy source in the usual sense) or that the vacuum furnishes energy ("regauging", Bearden). The first is conventional physics, limited to a few joules; the second would amount to a violation of energy conservation as accepted. Bearden's "magnetic field missing around the flux path, A potential present" statements (pp.5, 17) are given with no numbers.

Verdict (claim 3, split)

• Apparent COP above 1 measured on a scope (open loop): Tested, not supported as real power. The readings come from a math-channel RMS that the instrument maker's support staff call meaningless, the inventors' own two figures disagree by 3.54 against about 2.2, the load was not hot enough, and the effect appears only with a conditioned non-linear load. I weigh Naudin's candid notes ("apparent", "measurement artifacts remain possible", "pessimistic about calorimetric tests") as evidence in the same direction.
• Sustained COP above 1 / self-running closed loop: Never properly tested. No closed-loop run exists in 8 years of effort by Naudin; no calorimetry. Energy conservation makes the strong form very improbable, but the review should not call it "Refuted by calculation": the Jelinek-Pekarek proof is valid only for lossless anhysteretic cores and has errors in detail. The finite stored-energy route is not excluded and would show up as a transient COP that decays; no one reports one.

6. Claim 4: Mizuno-Ohmori plasma electrolysis reactor gives reproducible excess heat

a. Mathematics (Naudin's accounting)

Naudin's medium-power run: input 289.2 W for 82 s = 23,714 J; output = 10 mL evaporated x 2260 J/g + 150 mL x 21 K x 4.18 = 22,600 + 13,167 = 35,767 J; ratio 1.51 [read, calc]. The two terms are right arithmetically. The issue is what the 10 mL and the 21 K represent. If the mass loss is read after the cell has cooled (EarthTech says Naudin does this), the heat stored as a temperature rise has been counted once as sensible heat and again as the heat later used to evaporate water. If all 10 g were evaporated within the 82 s of power-on, evaporation alone would absorb 22,600/82 = 276 W, almost the whole 289 W input, on top of 161 W for heating, which would itself be the "excess". The page does not say when the cell was weighed [read]. The calibration (electric heater, 81 to 96 C, 8 mL, 0.985) shows the method works for a heater run in the same beaker, but the calibration used a different input meter (EKM 265 power meter) from the plasma runs (Fluke 123 scope and CIE CA-60A clamp, DC accuracy +/-1.5 %), and it measured a 190 s run with 600 mL, not the 82 s run with 150 mL [read]. A calibration with the plasma instrumentation and an equal evaporation fraction was not shown.

b. Consistency with experiment

• EarthTech (Little and Luce, 15 Jul 2003), direct replication on Naudin's own method with a Clarke-Hess 2330 wattmeter (a power meter with a "proven track record on a wide variety of waveforms"), a continuously logged balance (10 mg), 15 runs in a Dewar: COP2 about 1.00, average 0.98. Their "COP with cooldown", computed Naudin's way (adding the energy of the evaporation after power-off), ranges from 1.27 to 1.77 in their Table 2 [read: research/pages/et_replicationjln_2003.txt]. Naudin's typical figure, which they quote, is 1.6 (sometimes above 2.5). So EarthTech reproduced Naudin's numbers (1.3-1.8) with their own apparatus by using his accounting, and got 1.0 without it. That is a strong result: it is the same observation as Naudin's, shown to follow from the accounting convention. I note two limits: their cell was 600 g in a Dewar, not Naudin's 150 mL "Tiny CFR" or beaker; and the statement that Naudin adds the cooldown evaporation comes from a private communication with Naudin [read].
• Kowalski, Little and Luce, ICCF-12 (2005): in a Mizuno-type cell, mean COP 1.00 +/- 0.02 and 1.03 at 350 V, "none ... significantly different from 1.00"; they name unnoticed splashing and foam droplets as plausible sources of an illusory voltage-dependent excess [web: https://lenr-canr.org/acrobat/KowalskiLsearchingfa.pdf, Investigator]. They report Naudin's CFR1.1 COP as 1.85 at 150 V and 2.5 A.
• Mizuno 2004: "there have been few claimed replications of excess hydrogen, even in rare cases in which excess heat is claimed. In most cases, no excess heat or hydrogen is observed" [web: https://www.lenr-canr.org/acrobat/MizunoTgenerationa.pdf, Investigator]. This contradicts Naudin's "fully replicable".
• Positives by others that I will not suppress: Mizuno and Ohmori, original claims (200 % of input in some runs; Jpn. J. Appl. Phys. 39, 6055, 2000, via Naudin's page); Cirillo, Dattilo and Iorio (ICCF-11, 2004) COP 1.31-1.41 at 350 V (the paper Kowalski et al. tried to replicate); Mizuno and Toriyabe's explosive event in a 0.2 M K2CO3 / tungsten cell with an estimated heat 800 times input [web: Investigator's lenr-canr note]. None has been shown to include a calorimetric control against droplet entrainment and post-run mass accounting.
• Naudin's own low-power run in the same set-up gives 0.988 and the heater calibration 0.985, which is a modest internal control showing his method is not crudely biased at low power [read]. The medium-power claim then sits at the point where steam production is strongest, which is exactly where droplet carry-over and a mass-loss accounting error would be largest.

c. Tested? By whom? Replicated?

Tested by EarthTech (twice: 1998-2000 on Mizuno's own request, failed; 2003 on Naudin's, failed), by Kowalski / Little / Luce (2005, failed), by Naudin (positive, his own accounting), by Cirillo et al. (positive, not reviewed here). Naudin's 33 runs include "a blank run" per Kowalski; the run table is an image I did not read [read]. Kowalski's 2003 reaction to Naudin ("impressive ... reasonably reproducible ... skeptical accusations raised about Mizuno can also be raised about Naudin's data") predates the EarthTech result [read]. The rejection was by test (EarthTech built the experiment and offered Naudin's own convention a column), not by dismissal; Naudin's reply to EarthTech is not found [web: Investigator].

d. What would have to change

Excess heat from a tungsten cathode in K2CO3 on the scale of 1.5-3 times the input would need a nuclear or chemical energy source with no matching radiation, ash or hydrogen deficit. Naudin's explanation, "atomic hydrogen trapped in the incandescent tungsten ... a real cold fusion effect", is a mechanism statement with no measurement of neutrons, helium or transmutation. Standard nuclear physics gives no route at these energies.

Verdict (claim 4)

Tested, not supported for Naudin's reactor: independent instrumented replication gives COP 1.00, and the same data reproduce his reported 1.3-1.8 when the cooldown evaporation is added. The more general Mizuno plasma-electrolysis excess-heat claim is Never properly tested in a form that excludes droplet entrainment and post-run accounting, with a long record of non-replication by Mizuno's own account; I do not extend Naudin's negative to Mizuno's own experiments, which I have not reviewed.

7. Claim 5: Moller's Atomic Hydrogen Generator gives COP 5 to 21

a. Mathematics

Naudin's method [read: Investigator's saved pages]: output P_out = flow (m^3/h) x dT x 1161.11 W (4.18e6 J/m^3K / 3600 s/h) [calc: 4.18e6/3600 = 1161]. Input: "Rms Voltage and Rms Current" for the half-sine pulse runs; "average voltage and current measured on the digital oscilloscope" for the 5 % duty runs. The filament temperature formula T = (P/(kS))^(1/4) with S = 1e-3 m^2 treats the whole input as radiated; for a 0.25 mm filament the actual surface is much smaller, but I did not recompute his table.

A specific hazard I want checked. For a rectangular pulse train of peak voltage Vp, peak current Ip and duty D, the true mean input power is D Vp Ip. The product of the averages is (D Vp)(D Ip) = D^2 Vp Ip, which is smaller by the factor D. If the 5 % duty-cycle input was formed from separately averaged voltage and current, the apparent COP is too high by 1/D = 20, and a true COP would be the apparent COP times D: 11.53 x 0.05 = 0.58 for the 1153 % run and 21 x 0.05 = 1.05 for the COP 21 run [calc]. Naudin's statement that efficiency increases "up to 1153 % while the duty cycle decreases up to 5 %" is the pattern this artifact would produce (apparent COP proportional to 1/D). The same page elsewhere says power was measured with the Fluke 123 oscilloscope "so as to get a more accurate measurement of the electrical power", which may mean a proper sampled v x i average. The pages do not say, and the data table is an image. I am not asserting the artifact happened; I am stating a test the data can pass or fail: tabulate apparent COP against duty cycle for all pulsed runs and see whether COP x D is constant. That check needs only Naudin's MAHG data table, which the Investigator listed as not yet read.

b. Consistency with experiment

• Naudin's pure-DC runs 38-40 (the "blank") gave 64 %; Naudin explains the missing 36 % as losses, with the filament only at 774 K. The pulsed runs 11-37 gave 224 % average (net gain 464 W) and the 2 h runs about 240 %. The Frolov-Moller report that Naudin quotes shows the same split: 82 % with smoothed pulses, 124-130 % with rectified pulses, about 250 % with half-period diode pulses, "with the remark that this gave reason to think that a possible mistake had been made" [read: Investigator, mahg1.htm; Frolov's own warning]. The pattern, "pulsed good, DC bad, with the same average heater power", is what an input-power measurement error for non-sinusoidal, non-constant current would produce.
• The three input instruments agreeing within +/-2 % (analog meters with a shunt, Fluke 189 with a VC601 clamp, scope with the clamp) [read] is a fair control, but the report gives their agreement for the logged "RMS" input, and I cannot tell whether it was run in pulsed mode. A test that would settle it was never done: replace the tungsten filament with a pure resistor in the same pulsed circuit and run the same flow calorimetry. If the calorimeter reads the pulsed input power correctly the apparatus is calibrated; if it reads 3.5 times too high the claim is explained.
• Kowalski, a physicist, writes that any chemist would say the dissociation / recombination cycle cannot yield net energy and asks where the energy comes from [read]. The thermodynamics are simple: recombining H atoms into H2 returns exactly the dissociation energy that the tungsten filament supplied, since the cycle is closed [memory, high confidence]. Langmuir's atomic-hydrogen torch concentrates heat but does not create it. The only difference a cycle can make is where the heat is released, and the calorimeter takes it all anyway.
• Naudin states there were no anomalous radiations (a Gamma-Scout counter beside run 87) [read]; neither fusion nor fission radiation is claimed.

c. Tested? By whom?

Naudin (his own calorimetry) and Frolov / Moller (Faraday Lab, the originators, 2003). Moller and Frolov hold a financial stake (50-50 contract, Frolov quoted in the PESN article and by Kowalski) [read]. No neutral laboratory. A planned test in Australia was announced by Frolov [read: Kowalski item 289]; no result found. The Investigator did not reach New Energy Technologies no. 3 or the arizonaenergy.org page [read: sites.md].

d. What would have to change

A closed hydrogen cycle returning more heat than was put in would violate the first law. The authors invoke zero-point or "negentropy" energy (Moller reportedly thinks empty space supplies "free energy" [read: Kowalski]).

Verdict (claim 5)

Never properly tested. The numbers are the originators' and Naudin's own, rest on an input-power definition for pulsed non-linear loads that is either stated vaguely ("average voltage and current") or as RMS for half-sine pulses, and the one control (DC blank) changes the supply as well. The 1/D signature above is open. If the dummy-resistor control reads high, the verdict becomes Tested, not supported by the inventor's own apparatus. Energy conservation makes the strong form (COP 5-21 sustained) very improbable but the claim is not refuted by calculation until the input measurement is shown to be right.

8. Claim 6: two Cornille-Naudin patents

FR 2 789 818 A1 (fast charger)

a. Mathematics. The patent states the textbook half-energy loss when a capacitor is charged from a fixed voltage through a resistor, and proposes stepwise charging, which is physical [read]. A switched series-inductor charger is a buck-type converter (the deep reader's description; [memory] high confidence) and can have efficiency above 90 %. The patent's numbers are loose (30 s against 83 s is 2.77, quoted as 2.5 and as "divided by 3") and no efficiency definition or measurement is given. b. Experiment. Within normal engineering; no overunity claim. c. Tested. One unscaled voltage-time figure, no input-energy record. The French examiner cited five documents against claims 1, 2, 4 and 5 (Schneider Electric X; Braun, Matsushita Y; two Philips A) [read: deep]. d. Change in physics: none. Verdict: Plausible, untested (as a more efficient charger; it is ordinary switched-mode technique with nothing shown beyond one curve).

US 2005/0057116 A1 / WO 02/075912 / FR 2 822 590 ("efficiency of more than one")

a. Mathematics. The base step is correct: charge a capacitor, then reduce its capacitance at constant charge, so V rises by beta = C1/C2 and the stored energy (1/2)Q^2/C rises by the same factor. I recomputed C = 3.54 nF, C1 = 1.77 nF, beta about 594 for 300 mm and epsilon_r 4, the 0.175 J figure, and 338 % from beta = 200 [read: deep, calc]. Problems: - The patent's own mechanical analogue says the gain is the work done by the person who withdraws the plates and "the efficiency of the operation cannot exceed 100 %" [read: [0004]]. In the plasma version the same work is called "almost free" because it is done over a few Debye lengths [0031]-[0032]. Then it gives three unreconciled sources for the extra energy: Coulomb forces over Debye lengths, "the surrounding magnetic ether" (the stated reason for 338 %), and the Earth's potential energy of "several millions of volts" [0109]-[0110]. None appears in the claims [read: deep]. - The multi-cycle formulas (22-28, growth like beta^3 and beta^(2n-1)) require recreating C1 "charged with the voltage V2" with no source connected. Conserving the charge gives voltage V1, not V2; the extra charge (beta - 1) Q is not accounted for [read: deep, section 9]. - Formula 19's "about 136 Torr" does not reproduce (about 34 Torr from the printed inputs) [read: deep]. - The only measurement is one voltage trace (Fig. 8, 1 Gohm probe): 500 V to 1750 V when K1 opens, a ratio of 3.5. That is a measurement of the capacitance change, not of energy. The patent reports no input energy (the 15 kV plasma ignition supply, the 2 kV bias, switching losses), no output energy, no repeats and no instrument model beyond "VHV probe HVD100" [read: deep]. The patent's own caveat gives "a resultant beta of 10" and "total efficiency may become smaller than one" [0064]. b. Experiment. Variable-capacitor electrostatic machines are old and work; the claim that this switch is nearly free is not tested. c. Tested? By the inventors, one trace. The history: the French application was granted (B1), the US application was assigned to AES and abandoned on 27 Jun 2006 for failure to respond to an Office action, the European application (EP 1389360) was deemed withdrawn in 2008 for an unpaid 2007 renewal fee, and AES was struck off in 2021 after a last filing of accounts for 2015 with no employees [web: Espacenet, EPO Register, Pappers; Investigator]. I treat this as an anomaly to analyse. Competing explanations: (i) cost; (ii) a substantive objection the applicants chose not to contest (the PCT search report / written opinion, the USPTO Office action and the INPI B1 claims are the deciding documents and were "not reached" by the Investigator); (iii) lost interest after a negative test. Single-source note: the "abandoned" and "deemed withdrawn" entries come from the official registers, which are reliable on status but silent on motive. d. What would change. A real efficiency above one needs the energy released in the extinction of the plasma sheet to be supplied by a source outside the measured energy balance. Conservation says the energy to remove the induced charges from the glass interface must be paid, by the plasma circuit; the patent does not measure it.

Verdict (claim 6, split): capacitance-switching voltage gain: Established (as a capacitor effect, and with the trace consistent). "Efficiency more than one" in the strong form: Never properly tested (no input energy, no output energy, one trace, patent withdrawn). The multi-cycle gain formulas: Internally inconsistent (charge not conserved in the recreation of C1, no source for the added charge) as written. The mechanical analogue in the same document says no efficiency above one.

9. Claim 7: other devices Naudin hosts or replicates

This is a set, not a single claim. I judge each, with its source.

A rigour note for the inventor: across these pages Naudin repeatedly publishes "apparent" or "light output" figures and then discloses in the same breath that closed-loop or input data are missing. That is honest. It also means his claim of "overunity" is rarely backed by his own numbers.

10. The Investigator's findings, weighed

Kept in the record whatever my verdict: - CFR: the EarthTech replication, Kowalski et al. 2005 and Mizuno 2004 (decisive for claim 4). Naudin's calibration and low-power pages (0.985, 0.988) are an honest internal control. - MEG: Bishop's critique, now obtained (an advocate's stance: "expose a fraud"; formula error corrected by Smith), Flickinger's note on non-linear loads, Bearden's 7.3 and the unfound replications, and the MEG patent's fee history. - MAHG: the pulsed vs DC pattern in Frolov's own report and the changed input method at 1153 %. My 1/D check is new. - Patents: the legal history and AES's end. The WIPO / USPTO / INPI documents are the decisive records not yet read. - Lifters: NASA MSFC, ISR and Campbell patent; Tajmar and Schreiber 2020; Porcelli et al. 2016; Buhler et al.; Cornille 2020. The last four are related work and all keep their place: two nulls, two positives with different mechanisms. - Later Naudin work not in the folder (BiTT / Thane Heins, ReGenX, GFS UAV): leads only; not reviewed.

11. How the claims were rejected

• Lifters in air: not rejected, absorbed. The mainstream explained them (ARL, ISR, Ianconescu) and Naudin's lifters are accepted as working. What was rejected is the extra mechanism, by tests (Tajmar, ISR) that have their own limits (single run, low sensitivity).
• Vacuum thrust: the strongest "rejection" is a combination of an undocumented Air Force null (Talley) and under-sensitive NASA-related nulls. ARL itself asked for a vacuum program in 2003. I found no record that a neutral, sensitive vacuum test with outgassing control was ever funded. That is a gap, but it is also true that the positives never came with controls. Fair summary: ignored rather than refuted; the nulls are not conclusive.
• CFR: rejected by a replication that gave Naudin's own accounting a column; fairest "rejection" in the folder.
• MEG: rejected by calculation (Bishop, Jelinek-Pekarek) and by a Tektronix support note, not by a neutral experiment. No neutral experiment was offered or run.
• MAHG: not rejected by anyone; ignored. The AES and Magnetic Energy companies went dark; the explanation space (commercial failure, secrecy, quiet negative tests, lost interest) cannot be decided from the folder.
• No evidence of blocked publication or of omitted or reanalysed data was found; Naudin himself published adverse notes (the Magnetic Gate, the Larsko "not a COP", MEG artifact). Smith's reply to Bishop was hosted by Naudin; the Czech critique and the Bishop paper were not.

12. Strongest honest case for Naudin

1. He publishes nearly everything: schematics, parts, videos, adverse results, and the prior art that undercuts him (the Magnetic Gate correction; "NO overunity here" on the magnetic amplifier; the Orbo "not a COP"). That is the opposite of concealment.
2. He ran a heater calibration and a low-power control for the CFR and a DC blank for the MAHG; a three-instrument agreement for the MAHG input.
3. The lifters work, scale to 130 W hovering craft, and one of the largest bodies of builders anywhere verified them; his pages helped bring government laboratories (ARL, Dstl, NASA-related groups) to study them.
4. In vacuum, ARL (an Army laboratory) considered the question open in 2003, and the question was never closed by a sensitive controlled test. Several groups report forces in vacuum.
5. For the MEG, the same high-voltage, non-sinusoidal waveforms that make scope math unreliable also make a negative claim hard to prove without a calorimeter; no one ran one.

13. Where the inventor's own work fails my standard

1. CFR: the accounting counts evaporation after power-off in addition to the temperature rise (per EarthTech), and the calibration used a different input meter. "Fully replicable" is contradicted by Mizuno's own 2004 statement and by two instrumented replications.
2. MEG: "apparent COP" taken from a scope math channel (RMS of power) that the instrument maker calls meaningless; a conditioned non-linear load that is required; a missing heat balance rescued with an untested "radiated in EM form" explanation; and the claimed COP of 7.3 is Bearden's, not Naudin's.
3. MAHG: ambiguous definition of input for 5 % duty runs (see section 7), no dummy-load calibration, "COP 21" labelled under two versions, and the "blank" at 64 % shows the calorimeter loses 36 % under DC but is assumed to lose nothing under pulses.
4. Lifter tube tests: the electrostatic pull on a 0.13 g foil is the same order as its weight; the conclusion "proves definitely" the ionic wind is not responsible goes beyond the data.
5. The Gravitec "electrogravitic" conclusion is drawn from a video Naudin did not make, from a chamber record he does not show, on a thrust "weaker than in air" and unpublished for four years.
6. Patent: the "efficiency of more than one" is asserted from a capacitance ratio; the energy accounting omits the ignition supply and the extinction work.

14. What would settle each part (suggested for the arbiter)

• Lifter vacuum force: a torsion-balance test of one asymmetric capacitor and its mirror (or a symmetric control) from 760 Torr to 1e-6 Torr, with bake-out and residual-gas analysis, measured at 10-50 kV with the apparatus on a fixed insulating pedestal inside a grounded cage; publish force against pressure and current.
• CFR: weigh the cell and log temperature continuously through cooldown; compute input, heat stored and evaporated once each; run a pure resistive heater in the same cell at the same evaporation fraction; add a calibration with the plasma instruments.
• MEG: a power analyser or sample-by-sample mean of v x i, a calorimetric check on the load, and a closed-loop run of at least some minutes with a logged battery voltage.
• MAHG: a pulsed resistive dummy heater in place of the filament; tabulate COP times duty cycle.
• Patent: measured energy in (including ignition) and out for 1000 cycles, with a calorimetric or integrated-power record.

15. Summary table