Correa (PAGD / "XS NRG" / Aether Motor): Arbiter's report

Inputs reconciled: facts.json, claims.md, physicist.md, mathematician.md, contrarian.md. Source folder: source folder: Correa (read only).

Tags: [read] = in the Correa printout (rexresearch.com compilation, text layer re-extracted by me with pdftotext); [calc] = my own arithmetic on [read] numbers, run in Python; [web] = URL; [memory] = recalled, with confidence.

Source gap (repeated from Step 1, still true)

Only one of the three files has a readable text layer: "Paul & Alexandra Correa _ Pulsed Abnormal Glow Discharges (PAGD).pdf" (a web-page printout of 44 pages). The two patent PDFs (US 5,449,989 and US 5,502,354) are image-only scans. Their body text was taken from the same patents as reproduced in the printout. US 5,416,391 exists only inside the printout. Not obtained: the original measurement records, the Labofex Development Report S3-001 in full, the 1996 Denver symposium paper, any Aether Motor documentation, and any independent test report. Every number below is the inventors' own. Where a conclusion rests on [memory] alone it is marked as such.

Web-search note: my own web-search budget was exhausted by the session limit in the first pass, so I could not re-run the "independent replication" search then. I relied on the three reviewers' searches, all of which found only the Correas' own pages. In the QE revision (Round 2 web checks below) I read the 2007 rohnermachine.com PDF and made further fetches; the search tool was again unavailable, so "no independent test found" still means "not found by three reviewers plus the limited checks below", not "none exists". I made four page fetches in the first pass (listed at the end).

Headline verdicts

Tested / never tested / dismissed (kept separate, as the contrarian requires)

• Tested and failed (neutral test): none on record for any claim. No reviewer found a published null replication.
• Tested by the inventors only, never independently: claims 1, 2 (one reaction-balance number), 3 (all four battery-accounting methods and the oscillographic per-pulse data), 5 (motor and rope-and-pulley torque tests).
• Never tested at all: ion energy (no retarding-field or time-of-flight data), local surface field (no field-enhancement measurement), closed-loop self-running, calorimetry of the whole circuit, Aspden-versus-Lorentz discrimination, the Aether Motor.
• Dismissed without testing: no document in the folder, and nothing the reviewers found, records a refusal to test or blocked publication. The only documented obstacle is Axelrad's note that the lab was closing for lack of funds [read]. So this is neglect plus a funding failure, not a documented suppression story. The press-release rhetoric about being ignored is the inventors' own [read] and is uncorroborated.

Recomputation where reviewers differed

A. Table 7 efficiency (physicist vs mathematician vs contrarian)

Disagreement: the mathematician and contrarian both found that the "DP Watts" column is (X - Y) x I, not X x I; they differ on row 14 (93% vs 67%) and the median (about 92% vs about 89%). The physicist did not recompute.

I re-read the table text [read] and recomputed all 14 rows [calc]. Notation: X = drive-pack DC volts, Y = DC volts across the plates, I = drive-pack amps, CP = charge pack.

• Printed DP watts = (X - Y) x I for rows 1-8 and 10-14. Row 1: (562 - 350) x 0.65 = 137.8 W, matches. The text calls this column "drive pack total watts output" [read], which it is not if Y is subtracted.
• CP watts = CP volts x CP amps for all rows except row 14. Row 14 prints 257.7 W but 263 V x 1.36 A = 357.7 W, and the printed 228% only works with 357.7 (357.7 / 156.8 = 228%). So 257.7 is a typo. The mathematician is right; the contrarian's row-14 figure of 67% used the typo. Correct value on a terminal-power basis: 357.7 / (549 x 0.70) = 93%.
• Row 9: DP prints 112.45 W, which equals (563 - 390) x 0.65, but the DP amps column says 0.75; CP prints 280 W = 373 V x 0.75 A while the CP amps column says 0.65. The two current entries look transposed. Efficiency on terminal power is 57% or 66% depending on which is right; either way below 100%.
• Row 10: printed 530%, but 387.3 / 67.2 = 576%. Row 9 printed DP is 112.45 but 129.8 W if I = 0.75.
• Efficiencies if the input is the drive-pack terminal power X x I (CP watts / (X x I)), rows 1-14: 128, 78, 74, 104, 95, 91, 80, 66, 57 to 66, 146, 99, 87, 116, 93 percent. Four of 14 rows exceed 100% (rows 1, 4, 10, 13). Median about 92%. Range about 57-146%. The printed 218-563% requires the (X - Y) convention.

Who is right: the mathematician on the numbers. Whether (X - Y) is a legitimate input is a topology question that nobody could settle from the Fig. 12 scan. If X is the pack's terminal voltage and I the current it delivers, the pack supplies X x I regardless of where the voltage is dropped. That is the natural reading, and it removes the gain. But the printout defines "PAGD V = (X-Y)/I_av" (units of ohms, not volts, as printed [read]; the table uses (X-Y) in volts), so the authors may intend a specific meaning of Y that the text does not state. I therefore do not say "Refuted by calculation" for claim 3: that would require the topology.

Further agreed points (reviewer calculations I accept after spot-checking the arithmetic): - Methods disagree with each other. Table 8 group 2 has the drive pack losing 7.3 Wh in 18 minutes (about 24 W) [read], while Table 7 drive currents of 0.35-0.75 A at about 560 V imply roughly 200-420 W, i.e. at least 59 Wh in 18 minutes [calc; different runs, so a pointer rather than proof]. Table 5's 0.988 kWh gained in 21.5 min is about 2.8 kW average, against 210-470 W charge-pack power in Table 7 [calc]. - The authors themselves say the open-circuit-voltage method "in all likelihood, overestimates", that methods can differ up to 10-fold, and that battery capacity is not invariant [read]. - Table 8 (pre- and post-run resistive discharge curves) does not depend on the (X - Y) convention, and it is the inventors' preferred method. So the Table 7 problem does not by itself dispose of the claim. What is left is battery accounting of a few percent of capacity, with lead-acid relaxation and hysteresis unresolved (the inventors report "memory effect" and 50-60% of rated capacity [read via physicist]). That is exactly why the claim is "Never properly tested" rather than "Refuted".

B. Ion energy and mass (physicist vs mathematician vs contrarian)

• Printout: 86.4 J, 5.8e-8 g Al (from 3.1e-8 cm3 at density 1.86 g/cm3), 1.3e15 ions, "4.8e5 eV", 1.7e8 cm/s, 143.6 dyn [read].
• 5.8e-8 g / 26.98 u = 1.29e15 ions [calc]. 86.4 J / (1.29e15 x 1.602e-19 J/eV) = 4.15e5 eV [calc], not 4.8e5. The mathematician and physicist saw this; 4.15e5 reproduces the quoted 1.7e8 cm/s, so 4.8e5 is a slip or uses unstated inputs. Immaterial to the conclusion.
• Density: aluminium is 2.699 g/cm3 [web: https://en.wikipedia.org/wiki/Aluminium]. The printout's 1.86 g/cm3 is not aluminium (the text mentions "Alzak" plates; whether 1.86 is an oxide layer is unknown). With 2.70 the numbers become 8.4e-8 g, 1.9e15 ions, 2.9e5 eV: same order, conclusions unchanged.
• It is circular: the ion energy is the claimed pulse energy divided by the ion count. All three reviewers agree. It is not a measurement.
• Energy per charge: the same text says reactor pulse voltage is 100-500 V and cannot exceed breakdown [read via physicist]. At 560 V a singly charged ion gains 560 eV; an Al13+ ion (the physical maximum) gains 7.3 keV. Total for 1.3e15 ions at 560 eV: 0.10 J; even for 13 charges: 1.5 J [calc]. Claimed: 86.4 J. A factor of 60 to 800 short.
• Energy to ablate: Al heating to boiling plus fusion plus vaporisation is about 24.2 x 2443 + 10.7e3 + 284e3 = 354 kJ/mol, or 13.1 kJ/g [calc from web values: https://en.wikipedia.org/wiki/Aluminium]. For 5.8e-8 g that is 7.6e-4 J. The claimed 86.4 J is about 1e5 times larger. The mathematician's 6e-4 J (from an unverified 10.5 kJ/g) is confirmed to within 25%.
• Measured ion energies: cathode-spot ions in vacuum arcs have directed kinetic energies of about 19-150 eV and velocities of order 1e4 m/s [web: https://www.osti.gov/servlets/purl/810482, Anders review, read by me]. The claimed 1.7e6 m/s (1.7e8 cm/s) is about 170 times faster. Ions in arcs can exceed the arc voltage (tens of eV from a ~20 V arc), so the electrostatic limit is not absolute, but nothing reaches 1e5 eV.
• Disagreement on label for "0.5 MeV": physicist "Refuted by calculation", mathematician "Internally inconsistent", contrarian "Never properly tested". I choose Refuted by calculation. The number is derived (not observed), so "never tested" would suggest it is an open measurement; and the energy bookkeeping above fails with accepted laws applied correctly. The internal inconsistency (voltage ceiling stated in the same text) is real and is noted.

C. "100x vacuum arcs" (mathematician only; others did not recompute)

The k values in Table 15 (force over i squared, Graneau form F = (mu0/4pi) k i^2) use the DC input current of 0.5-1.7 A [read via reviewers]. The text says pulse currents inside the reactor are 0.5 to over 200 A [read via mathematician and Step 1]. With i = 150 A: (1e-7)(150)^2 = 2.25e-3 N per unit k; 143.6 dyn = 1.436e-3 N gives k = 0.64 [calc, I reproduced 0.638]. So the "100x" disappears when normalised by a pulse current, if the peak current really is of that order during the force pulse. It depends on that assumption, so I label the result conditional. The force-from-momentum step also needs "m" to mean mass per pulse divided by the 40-50 ms pulse time: 9.86 g cm/s / 1.39 / 0.049 s = 145 dyn [calc]; as printed (F = mV/1.39) the units are wrong.

D. Fowler-Nordheim and Paschen (physicist vs mathematician vs contrarian)

• Mathematician's Paschen estimate for air at 0.8 Torr x 3 cm = 2.4 Torr cm gives 426 V [calc, I reproduced 426.4 V with A = 15, B = 365, gamma = 0.01]. Observed breakdown is about 500-560 V [read]. Same order; the constants are approximate and the tube is not parallel-plate. This shows gas (Townsend) breakdown can account for the operating voltage; it does not "refute" an anomaly claim. Hence I do not adopt the mathematician's "Refuted by calculation" label.
• The inventors compare the gap-average field (5,000-30,000 V/m) with the field Fowler-Nordheim needs (over 2e9 V/m) [read via physicist]. Fowler-Nordheim concerns the local field at the emitting surface; microprotrusions and adsorbates routinely enhance it by large factors [memory, high; all three reviewers]. The comparison is not like for like, so no violation is demonstrated, and no prediction distinguishes the "anomalous" case from enhancement. That is "Not testable as stated" (physicist), better than "Never properly tested" (contrarian), because nothing discriminating was claimed.
• The worked tubes run at 0.09-1.5 Torr, a gas discharge, where Fowler-Nordheim is not the controlling process. The only sealed, ~1e-4 Torr cases get no calculation in the text (mathematician); these stay "Never properly tested".
• The mathematician's point that the stated beta of 1e6-1e7 cannot be reproduced (the ratio is 1.3e5) is an additional internal inconsistency, minor.

Claim by claim

Claim 1: self-pulsing tube, emission below Fowler-Nordheim

• Agree: relaxation oscillation of a DC-fed, ballasted, capacitor-shunted discharge is textbook (neon-lamp oscillator) [memory, high; all three]. Nobody but the Correas has tested this specific tube. Contrarian credits an orderly dataset (14 configurations, systematic variation with pressure, plate material and ballast; break-in over more than 2e6 pulses [read via contrarian]).
• Disagree: label for the "anomaly" part (see D).
• Stronger argument: physicist and mathematician. The anomaly premise is a category error, and ordinary field enhancement plus gas breakdown suffices.
• Final: weak form Established; strong form Not testable as stated; sealed low-pressure case Never properly tested. If the strong form were true, little in mainstream physics would change (surface and gas-phase triggering physics only).
• Would settle it: measure the local emission (field-emission microscopy or a profilometer-based enhancement map) on the same plates at 1e-6 Torr, and compare the I-V curve with a Fowler-Nordheim plot including beta.

Claim 2: 0.5 MeV ions, 100-300 dyn, 100x arcs, Aspden's Law

• Agree: the arithmetic is recoverable; the ion energy is derived from the claimed pulse energy; no ion spectrometry reported; no independent repeat of the 245.2 dyn balance reading.
• Disagree: labels (B and C above). Physicist and mathematician called the force order "Plausible, untested"; contrarian "Never properly tested". I side with the contrarian: the 143.6 dyn figure uses the circular 1.7e8 cm/s velocity, so the momentum bookkeeping is not independent support, and the only direct number (245.2 dyn) is a single unreplicated, undescribed balance reading. A real recoil force from an arc is expected [memory, medium: Tanberg/Kobel recoil attributed to plasma jets; not web-verified by me], but the magnitude here is untested.
• Aspden's Law as printed: F = (qq'/r^3)[(v'.r)v - (m'/m)(v.r)v' - (v.v')r], with m'/m the ion-to-electron mass ratio [read]. The mathematician notes it lacks the factor (mu0/4pi) or 1/c^2 and so is not dimensionally a force as written, and the mass-ratio term makes the force depend on the particle mass, which ordinary field-based electrodynamics does not. The "recognition and experimental confirmation" cited in the printout are the inventors' own references (49-55) [read]; I could not verify them (Aspden's Wikipedia page returned 404; check failed). The inventors also note that Graneau rejects Aspden's reading of his own data [read]. The Aspden constant is a fit ("in the 10^4 range", contrarian). I rate the Aspden attribution Never properly tested: no neutral experiment discriminates it from Lorentz force plus plasma pressure.
• If true: the Lorentz force law and momentum conservation would need a mass-dependent, frame-dependent term, with the vacuum carrying the missing momentum.
• Would settle it: (i) a retarding-field analyser or time-of-flight measurement of the ion energy spectrum in the actual tube; (ii) a calibrated torsion or impulse balance on the same tube with a published calibration, pulse counting and independent repeat; (iii) force normalised by measured peak pulse current, to see whether k is about 0.6 or about 100.

Claim 3: over-unity circuit (218-563%)

• Agree: all four accounting methods are the inventors' own; the methods disagree; the authors admit overestimation and ignored delivery time; a closed-loop or calorimetric demonstration was not done. Nothing here is a documented neutral test, and no refusal to test is documented.
• Disagree: contrarian called Table 7 "Internally inconsistent" and the general claim "Never properly tested"; the mathematician called the general claim "Never properly tested" and noted the data do not support it; the physicist "Never properly tested". All three agree on the label for the general claim. The contrarian's row-14 error does not change the picture (corrected above).
• Contrarian's additional point that rms meters (30 and 100 kHz bandwidth) were used for pulsed currents is credible and unexcluded: an rms reading overstates mean current. The text does not state the meter mode, so this is an open artefact, not a finding.
• Per-pulse energies (Table 13/14, "3 to over 50 times"): derived from oscillograms across series capacitors (5.3 uF or 17.5 mF [read via reviewers]). 0.5 x 0.0175 F x (560 V)^2 = 2.7 kJ [calc], so 100-350 J pulses are circulating energy in a large AC-coupled capacitor; integrating v x i without a full-cycle window cannot give net energy. Many reactor-energy entries are "ND" (not determined).
• Strongest honest case (contrarian): four methods, consistent sign, published cell counts and caveats, and the test that would settle it is cheap. Strongest honest case against: the headline table does not survive recomputation, methods differ by 2-10x, the evidence is a few percent of battery capacity, everything is by the inventors with no blinding.
• Final: Table 7 as printed Internally inconsistent; net gain Never properly tested. Not "Tested, not supported", because no independent test was found (the 2007 rohnermachine.com PDF is a reprint of an Infinite Energy article, not a test; see Round 2 web checks); not "Refuted by calculation" for the general claim, because a real test of the whole circuit has not been done. (Table 7 as printed and Table 8 as an energy balance are separately refuted on the inventors' own numbers, see the QE revision.)
• If true: conservation of energy would need an unmetered source; thermodynamics and QED would have to be extended (inventors: "a further source of energy input").
• Would settle it: run the circuit from a calibrated DC supply (not a battery) with a wide-bandwidth true-mean-power analyser at both ports, or whole-circuit calorimetry, with a resistive load in place of the charge pack; run for hours; then close the loop (output feeding input) and see if it runs unattended.

Claim 4: surplus from zero-point or aether energy

• All three agree: no derivation. The text compares an energy density (86.4 J / 3.1e-8 cm3 = 2.8e9 J/cm3 [calc, reproduced]) with Aspden's vacuum value, a coincidence argument. The "dynamic capacitance 0.2-80 mF" for a 2 pF tube is derived by dividing the claimed energy by V^2, so it assumes the answer. The zero-point state is the lowest state; the Casimir effect is a boundary force, not a power source [memory, high; not web-checked, and no verdict depends on it]. Conditional on claim 3.
• Final: Not testable as stated. Would settle it: nothing separate; demonstrate claim 3 first, then show the surplus depends on a controlled environmental variable.

Claim 5: DC-to-AC inverter for motors (US 5,416,391)

• All three agree it is ordinary physics: a relaxation oscillator driving a motor, speed following pulse rate (synchronous speed = 120 f / poles). Nothing over-unity is required.
• Mathematician's unexplained point: runs 11-12 report 2800-3000 rpm at 10 pulses per second, far above synchronous speed (600 rpm for 2 poles at 10 Hz [calc]); the authors say the motor acts as an induction generator. Unexplained, not impossible with non-sinusoidal pulses.
• Final: mechanism Established; claimed performance Plausible, untested (no efficiency reported). Would settle it: an efficiency measurement at the motor shaft against true input power.

Claim 6: Aether Motor

• Agree: Only Axelrad's testimonial, who states no training in plasma physics [read]. No equations, circuit or data. "No external input" is not well defined (a ground and insulated bodies can collect atmospheric electricity, fair-weather field about 100 V/m [memory, high]).
• Orgone-accumulator controlled replications: the physicist and mathematician say none exist [memory, medium; not web-verified by me, check failed because search budget ran out]. No verdict depends on it, since the claim has no specification.
• Final: Not testable as stated. Would settle it: a schematic, build instructions and a metered, enclosed, input-free run.

Corrected claim list

1. A large-electrode, large-gap cold-cathode tube fed from DC through a ballast and shunt capacitance self-pulses (relaxation oscillation). Established. Whether its emission needs a mechanism beyond field enhancement and gas breakdown is unsupported and not testable as framed for the 0.09-1.5 Torr gas-discharge runs. The tubes sealed at 1e-6 to 2e-6 Torr (Penning-gauge value on the pumping system at seal-off, not re-measured in the sealed tube), and the claimed fall of breakdown voltage with plate area contrary to Paschen scaling, are never properly tested; Paschen's law does not reach them.
2. Each pulse ejects about 6e-8 g of cathode metal (inferred from crater volume using a density of 1.86 g/cm3, not aluminium's 2.70). The "0.5 MeV" ion energy is the claimed pulse energy divided by the ion count, not a measurement, and fails an energy-per-charge check (at most about 0.1 J, or about 1.5 J at the maximum charge state, against 86.4 J). One reaction-balance reading of 245 dyn is reported, unreplicated. The "100x vacuum arc" ratio comes from dividing by the DC input current (1.6 A) while the same text reports peak pulse currents above 150 A (internally inconsistent; QE deep read 4: k falls to the vacuum-arc value at 12-16 A and is about 0.6-1.1 at 150 A). The 86.4 J is untraceable to Table 14, and the printed 4.8e5 eV is 4.2e5 eV on the authors' own inputs.
3. A battery-fed, battery-collected circuit is reported by four accounting methods: Table 5 96-906% (plus a 6,750% outlier), Table 6 504-776%, Table 7 218-563%, Table 8 406-3,485%. Table 7's "drive pack watts" is (X - Y) x I, the power lost in the ballast resistor and input diodes, not the pack output (X x I); on X x I the runs are 57-146%, median about 92%, with arithmetic slips in rows 3, 9, 10 and 14. Table 8's drive-pack losses (5.6-89 W average) are below the 98 W minimum implied by the authors' own 0.2 A minimum current; group 6 on its own FIG. 20 currents is about 60-95%; its 3,485% does not reproduce. The 6,750% run has a mis-divided loss rate and a drive pack the method counts as nearly empty. The methods disagree among themselves by up to 10-fold (authors' own statement). No neutral test, closed loop or calorimetry was found (the 2007 rohnermachine.com PDF is a reprint of Infinite Energy articles relaying Correa data), so net gain remains never properly tested.
4. The explanation by zero-point or aether energy has no derivation or prediction beyond the claimed surplus.
5. The tube works as a pulsed-AC supply for induction motors; efficiency not reported; high-rpm run unexplained.
6. The Aether Motor is described by testimonial only; no technical content.

What would settle it

1. True mean input and output power, wide bandwidth, at both ports, from a calibrated DC supply into a resistive or calorimetric load, for several hours. If output stays at or below input, claims 3 and 4 fail; if above, send it to an independent lab.
2. Closed loop: feed output into the drive supply and see whether the tube runs without an external source.
3. Retarding-field analyser or time-of-flight on the ion beam: decides the "0.5 MeV" question directly.
4. Calibrated, repeatable impulse balance with peak-current measurement: decides force and the "100x" ratio.
5. Local surface field and enhancement factor measurement on the plates: decides whether any emission anomaly exists.

Open questions

• (Resolved by QE revision) What does Y mean in Table 7 and what is the circuit topology? Y is the tube voltage; the packs are coupled only through capacitors; "DP Watts" is ballast loss.
• Are CP/DP amps in Table 7 row 9 transposed, and is 1.86 g/cm3 the density of anodised ("Alzak") material rather than bare aluminium?
• (Resolved by QE revision) Why does Table 8's drive-pack loss differ by 10x from Table 7 drive power? Because Table 8 books too little loss (below the 98 W floor), and the FIG. 20 "drive wattage" appears to be ballast dissipation too (inference; the figure is not in the printout).
• Was the force of 245.2 dyn a time-averaged balance reading or a per-pulse impulse, and at what pulse rate? (The 143.6 dyn figure is divided by a 40-50 ms pulse time, while the balance reading's timing is not stated.)
• Is the pulse peak current of order 150 A at the instant of the force? The k calculation depends on it.
• Which meter mode (rms or mean) was used on the pulsed charge currents? Do the pulsed charge waveforms inflate apparent battery gain through relaxation and hysteresis effects? [lead-acid recovery effects: memory, medium; not web-checked]
• Is there any third-party test of the Correa apparatus (Aspden, Mallove and Axelrad endorsements are not tests; the 2007 PDF at rohnermachine.com, examined in the QE revision, is a reprint of a Mike Carrell article in Infinite Energy that relays Correa data, not a test)?

Web checks done

1. https://en.wikipedia.org/wiki/Aluminium: heat of fusion 10.71 kJ/mol, vaporisation 284 kJ/mol, boiling point 2743 K, molar heat 24.20 J/(mol K), density 2.699 g/cm3. Used for the ablation energy (13.1 kJ/g, 7.6e-4 J for 5.8e-8 g) and the density flag. Retagged [web].
2. https://www.osti.gov/servlets/purl/810482 (Anders, cathodic arc plasma deposition review, read by me): ion velocities of order 1e4 m/s; directed kinetic energy of 19-150 eV; multiple charge states. Retagged [web]. The reviewers' separate figure of Tanberg's 1.6e4 m/s for copper is consistent with this.
3. https://en.wikipedia.org/wiki/Vacuum_arc: no usable numbers; check failed to add information.
4. https://en.wikipedia.org/wiki/Harold_Aspden: 404; check failed. Aspden's standing in mainstream literature stays [memory, medium] and the inventors' printed claim of "recognition" stays [read], uncorroborated. No verdict depends on it.
5. Replication search: could not run in the first pass (web-search budget exhausted); reviewers' results stood. Redone in part in the QE revision (see Round 2 web checks). Lead-acid relaxation was web-checked in the QE revision (surface charge and rest time retagged [web] there). Not web-checked, still [memory]: Reich accumulator replication record (medium), Tanberg/Kobel recoil attribution to plasma jets (medium), Casimir/zero-point statements (high). None of these changes a verdict.

QE revision (2026-10-03, after qe.md verdict MINOR)

Tags as above; [deep] = the QE-commissioned deep reads in deep/ (Opus), [qe] = taken from qe.md, which cites the extracted text.

Source-gap update: the two patent scans (US 5,449,989 and US 5,502,354) now have OCR text for all pages (Step 0). The "image-only" statement in the Source gap section above is out of date. The drive-pack circuit sheets and Table 7 were read as page images in the deep read.

Which QE remediation items were done: deep reads 1 (circuit sheets, Table 7) and 2 (Tables 5, 6, 8 and the four methods) were run in the first QE pass. Round 2 then added deep read 3 (US 5,502,354 Examples 1-6 and vacuum processing, pp.29-35), deep read 4 (Tables 12-16, per-pulse energies, the worked example and the 245.2 dyn reading, pp.36-43) and a read of the motor patent examples (US 5,416,391 Examples 1-5, pp.11-13), all in deep/Paul-Alexandra-Correa-Pulsed-Abnormal-Glow-Discharges-PAGD-8687ea.md. Deep read 5 (transcription check of Tables 1-9) is not separately reported in that file; the arithmetic checks in deep reads 1-4 cover Tables 5-8 and 12-15 only. Three web checks were made in Round 2 (results in "Round 2 web checks" below); items in the text above that read "[memory, medium]" for lead-acid relaxation are retagged there, and the "not found by three reviewers" statement is updated there.

New evidence

1. Circuit topology settled [deep: FIG. 8-11, pp.5-8 page images; qe.md 3a]. In FIG. 9 and variants 10A-C the drive loop is one series loop: drive pack (DP), diode D1, ballast R1, tube, diode D4. The charge pack (CP) connects only through capacitors C3 and C5 (with the bridge and C7a/C7b); the authors state the aim is "DC isolation of the two battery packs" [read: p.18]. The triode variant 10B has one wire from the axial electrode to bridge node P, but it forms no loop. So there is no direct DC path from DP to CP.
2. Meaning of Table 7 columns [deep: p.34 image]. Column 8 "DP Watts" = (X - Y) x I in 13 of 14 rows; column 9 "PAGD Volts" = X - Y in all rows; column 14 "Total Resistance" = (X - Y)/I, about 230-330 ohms, which is the ballast scale (R1 is 300 or 600 ohms elsewhere), not the "entire circuit". X - Y = voltage across R1 and the two diodes. The printed formula "PAGD V = (X-Y)/I_av" has units of ohms and does not match the table. Y is the tube voltage. The drive pack really delivers X x I. Row 1: 365 W = 138 W in the ballast + 228 W in the tube.
3. Table 7 recomputed on X x I [deep; calc]. Ratios CP W / (X x I): 57-146%, median about 92%, with the V x A fixes in rows 9 and 14; 10 of 14 runs are at or below 100%; the four above are rows 1 (128%), 4 (104%), 10 (146%), 13 (116%). Scan check: the slips in rows 3 (col. 14 prints 243, recompute 320), 9, 10 and 14 are in the printed patent, not OCR errors. The diode-versus-triode ordering survives (about 100% versus 80% on X x I), the gain does not.
4. Table 8 is not credible as an energy balance [deep: pp.17-19, 26-27; page image p.26]. (a) The authors give 0.2-0.6 A from a pack at 489-590 V, so the DP must supply at least about 98 W; Table 8's booked DP losses average only 5.6-89 W (all six groups below the floor; group 3 at 5.6 W is below the 12 W the 300-ohm ballast alone dissipates at 0.2 A). (b) Group 6: the authors' FIG. 20 "drive wattage" of 36.6-57.8 W equals I squared x 300 ohm for I = 0.35-0.44 A, so X x I is about 190-240 W, 233-303 Wh over 74 min against 37.7 Wh booked; breakeven becomes about 60-78% (73-95% using FIG. 20's projected gain). That the FIG. 20 figure is a ballast loss is an inference; the figure is not in the printout. (c) The Wh figures are extrapolations along discharge curves to cutoffs of 115 W (DP, 489 V, 10.64 V/cell) and 90 W (CP, 274 V, 9.44 V/cell, below the manufacturer's 10.3 V/cell end point), started 15-30 min after loading, while the authors' FIG. 13 says voltage is still relaxing. Illustration: a 0.41 W change in starting load power moves the DP group-1 projection by about 32 Wh, the whole booked loss. (d) Each pre-run measurement discharge itself draws about 130 W x 1-3 h, 4-60 times the booked DP losses, and how "before" is defined is not stated. (e) Group 3's 3,485% cannot be reproduced (3,245% from printed Wh; 3,423% from printed Wh/h). Group 2 gain, group 3 loss and group 4 net also have small slips. The "406-3,485%" range is correct as a printed range.
5. Table 5 run 5, 6,750%, with the motor on [deep: pp.23-24; qe.md 3c]. It is the largest efficiency in the documents and was missing from the earlier review. Loss rate is mis-divided: 0.045 kWh / (80/60 h) = 0.034 kWh/h, not 0.016; corrected breakeven about 3,200%, and the CP gain is mis-subtracted (1.410, not 1.440). More basic: the open-voltage method rates the DP at 3.2% (0.101 kWh) at the start, which caps any loss it can register; 80 min at even 0.2 A and 517 V draws 0.138 kWh. The contrarian's "Table 5 gives up to 776%" was wrong: 776% is Table 6 run 1 [qe; read: l.2522]; Table 5 runs are 96% (run 6) to 906% (run 8), plus 6,750%. Table 5 run 6 books 1.551 kWh lost in 21.5 min (4.3 kW) through a 300-ohm ballast from 589 V (maximum possible about 1.2 kW), and run 1 books 2.8 kW into the CP (about 7.7 A, against a maximum of 1.36 A in Table 7) [deep; calc]. The open-voltage method (Tables 5) therefore is not an energy measurement. Table 6 (method 2) is internally consistent but implies 3.46 kW (about 10 A) into the CP in run 1.
6. Authors' own admissions [deep; read: p.17-19]. The methods differ up to 10-fold; the open-voltage method "overestimates"; the output energy "is not directly related to that utilized in triggering the pulses"; removing the C7 bridge suppresses PAGD; output rises with the series value of C3, C5 and C7 (20,700-34,700 microfarad, large unmetered stores). Adding to this [qe, read: digests Eng-Conv.01 pp.18, 20]: the CP cell count sets the oscillation (29 to 31 cells cuts the pulse rate 10-fold; more cells extinguish it), so the "output" is part of the oscillator; and at about 500 mA applied the pulse AC rms current is up to 7.5 times the DC current, with ">150 A" peaks from 1.2-1.7 A input. Current was read on Beckman rms multimeters, mode not stated [deep; read: p.16]. Both points support the earlier "rms meter artefact" open question: an rms reading is at least the mean current, so both DP and CP watts are overstated by unknown factors.
7. Sealed high-vacuum tubes [deep read 3, pp.29-35; read: PAGD.txt l.1097-1102, l.1220-1223, l.3108; digest 5502354.02 p.26]. Example 1 (128 cm2 plates, 5 cm apart) was sealed at 1e-6 Torr and showed PAGD from about 10 to 150 mA RMS with the plateau voltage about 850 V (the same tube gives about 668 V at 500 mA, p.35, so Vb depends on the supply as well as the tube). Example 5/10 tubes were sealed at 2e-6 Torr; 16 and 64 cm2 tubes did not respond even at 3.3 kV (Table 6: above 3.5 kV), while a 256 cm2 tube fired at 287 V (extinction 284 V, one tube). At 1e-6 Torr, pd is about 5e-6 Torr cm and the ionisation mean free path is about 1,200 m, so an electron crossing 5 cm makes about 4e-5 ionisations [calc, deep read 3]: gas avalanche cannot explain these, and Paschen scaling does not apply to them at all. The earlier "~1e-4 Torr" in this file was wrong. Correction to an earlier QE statement: the first QE pass said seal-off pressure "was not gauge-measured" (thermocouple gauge limit, sourced to a digest of the other patent). Deep read 3 shows that was a cross-document transfer: the PAGD text itself says the rig had "thermocouple and Penning gauges (for rotary and diffusion vacua, respectively)", that tubes were "a vacuum which measured 10^-6 Torr at time of seal off", and Table 6 column G is "the vacuum measured when the respective housings were sealed" [read: p.30, p.34; confirmed by me in _text, Penning gauge at l.3108]. The supportable caveat is: a Penning-gauge value on the pumping system at seal-off (factor-of-two scale uncertainty [memory, medium]); not re-measured inside the sealed tube, after cooling or after discharge running; gas composition unknown (residual argon, Silicone 705 oil vapour, which the authors themselves raise as a possibility for the diffusion-pumped tubes [read: p.31], and hydrogen or desorbed gas); no getter or residual-gas analyser described. Qualitative checks (Tesla coil "black vacuum", x-ray onset at 25 kV) were additional, not a substitute. The authors processed tubes at 25 kV and saw x-rays at the plate edges at 1e-6 Torr, so these tubes carry field-emission current at kilovolt scale; the "no response at 3.3 kV" tubes fit a steep emission threshold, not anything exotic.

High-vacuum Paschen scope (from deep read 3). Paschen's law applies in the 0.0038-0.7 Torr pumpdown runs (Tables 2-5, argon, 5.5 cm gap), and there the authors' plate-area data do contradict plane-parallel Paschen scaling as applied: at fixed voltage the pressure falls by factors 0.52, 0.44, 0.24 for plate-area ratios 2, 4, 8, and at fixed pressure the voltage falls to 91.7, 85.9, 78.7 percent of the 16 cm2 value [read: p.32-34]. Weaker than claimed, because the cheap explanations are not excluded: the 16 cm2 plates are 4 cm square against a 5-5.5 cm gap, so field lines run round the edges and to the glass wall (the effective path exceeds d); no edge-field, wall-charging or glass-envelope control is reported; the table statistics are triangulated from the same three curves, so the quoted SEMs understate the uncertainty; and the vacuum-insulation literature has a known area effect [memory, medium; not web-checked]. In the sealed 2e-6 Torr series the area effect is much larger (voltage falls at least 4.5x from 64 to 128 cm2, then 2.7x to 256 cm2, exponent n about 1.5 to above 2.2 against Paschen's 0), but Paschen is outside its domain there, so the "contradiction" is with a law that does not apply; the data are a one-tube-per-group set with different gaps, currents (200 mA to 1 A) and supplies. At constant pd the authors' own scaling fails too: group 1 (3.6 cm, 2.8e-6 Torr) did nothing above 3.5 kV while group 2 (5 cm, 2e-6 Torr) fired at 774 V, which the authors admit [read: p.34]. Verdict for the sealed tubes and the area effect: Never properly tested (no independent measurement), with these alternatives open. 8. Cathode as fuel, energy per gram [qe 3c; calc]. The authors call cathode erosion "a fuel consumption process" [read]; they cite 0.25 kWh per 1e4 pulses and 20-40 MWh per H34 cathode [read]. Using the claimed 86.4 J per 5.8e-8 g: 1.5e9 J/g, about 5e4 times the energy of burning aluminium (about 31 kJ/g [memory, high]) and about 2% of uranium fission (about 8e10 J/g [memory, high]). Either the 86.4 J is wrong, or the pulse is not driven by the cathode metal. Deep read 4 adds that 86.4 J cannot be traced to any row or natural mean of Table 14 (the nearest independent figure is 90 J per pulse from "0.25 kWh per 10,000 pulses" [read: p.40]) and is a net electrical figure (output minus input), not an ion kinetic energy. 9. Other items to carry into facts and primer [qe 3c]: the largest efficiency claims by table; people and events (Ernesto Manuel's 1969 PAGD patent US 3,471,316; A. Chernetskii's 1989 reactor; Kimblin; Plyutto; Compton; ABRI second lab; testimonials by Tiller, Carrell, Soudak); X-rays and "cathode rays" below 1e-6 Torr at higher voltage (a safety point for anyone building a tube). Digest exponents for the two scans are unreliable (OCR drops superscripts): true values 4.8e5 eV (not 4.8e9), H34 lifetime 1.2e9 pulses (not 1.2e8), k about 1e3-1e4 i squared versus 1e2 i squared; no downstream file used the wrong values. The digest's "230% to 6,750%" for Table 8 merges Tables 5 and 8.

Changed verdicts and reasons

Other corrections to the text above: Section A's statement that the topology "could not be read" is superseded by the deep read. Section D's "sealed ~1e-4 Torr cases" and the mathematician-beta point are superseded by item 7. The statement in Claim 3 above that Table 8 "does not depend on the (X - Y) convention" remains true as to the formula but Table 8 has its own flaw (item 4). Table 8 drive-pack accounting is unresolved only in the sense that its source curves (FIGS. 13-21) are not in the printout.

What would still settle it

Unchanged from the list above, with two cheap additions: (a) log X, I and CP current continuously and integrate X x I dt and V_CP x I_CP dt, with a dummy load in place of the tube [deep]; (b) measure pressure and gas composition (residual-gas analyser) inside the sealed high-vacuum tubes after running, and record breakdown voltage against plate area, with a calibrated surface-field map.

Round 2 (deep reads 3 and 4, motor examples, three web checks)

Tags: [deep] = deep/Paul-Alexandra-Correa-Pulsed-Abnormal-Glow-Discharges-PAGD-8687ea.md (sections "QE round 2: US 5,416,391", "QE round 2: US 5,502,354 Examples 1-6", "Second QE round (deep read 4)"); [web] with URL.

Web checks

1. rohnermachine.com/Files/CPAGD.pdf (2007) [web: http://www.rohnermachine.com/Files/CPAGD.pdf; text read by me with pdftotext from the downloaded file, 13 pages, PDF created 19 Nov 2007]. It is a reprint of Mike Carrell's Infinite Energy articles, "The Correa Invention: An Overview and an Investigation in Progress" and its follow-up "The Correa PAGD Reactor: Errata and Supplement". It is not an independent test and not a replication. Carrell states he had "several long conversations with Dr. Correa, access to the full text of three patents, and some recent data taken with digital instrumentation" supplied by Correa; the oscillograms were "furnished" by Dr. Correa and traced by Carrell, "the waveforms at the reactor itself ... remain the proprietary information of the Correas", and the 8-hour closed-loop battery-swap test is described only by the patent and by Dr. Correa's word ("Numerical data from the test is not publicly available"). What it adds: (a) Carrell himself says the Beckman rms meters "were probably not giving accurate readings because of the low duty cycle of the pulses", which supports the rms-meter open question; (b) Carrell concedes that Table 8 loss figures are "differences of large numbers" (the 6 Wh DP loss) and "vulnerable to measurement errors"; (c) the follow-up errata corrects Carrell's own mistakes in IE #8 Fig. 4 after Dr. Mitchell Swartz and Vortex-l readers found them (that is scrutiny of an article's arithmetic, not a test of the device); (d) Correa-supplied digital oscillograms (80 ms record, 0.5 pps) are said to show the drive pack delivering about 0.31 A for 25 ms with background current otherwise, against a 65 A pulse into the charge pack, and the energy out/in ratio 101 (445 J out, about 4.4 J in by my arithmetic from 570 V x 0.31 A x 25 ms [calc]). Point (d) bears on the deep-read-4 inference: if the drive pack really draws near zero between pulses, V x I x t_pulse is a fair input for those runs and the "omitted inter-pulse charging energy" explanation of the large ratios at low pulse rate is weakened (it remains conditional). It does not make the data independent: the traces are from the inventors, instrument calibration, bandwidth and the integration window are unstated, and the large pulse energies still sit on an AC-coupled 17.5 mF bank whose stored energy (0.5 C V^2) is of the same order as the "output". Source for Carrell as the author: the PDF's byline text "Mike Carrell" [read in the PDF]; the PDF metadata author is "Bob" (a site uploader, unverified).
2. Infinite Energy (Mallove) and Aspden: independent tests? The only Infinite Energy item I could read is item 1 above, which is a promotional-and-explanatory reprint (it says the patents and a Correa laboratory report were published in IE #7), so IE published the Correas' work; it did not test it. It cites Aspden's privately published Energy Science Report No. 8, "Power from Space: The Correa Invention", as an interpretation (aether strain), not a test. My search tool was unavailable (budget exhausted) and attempts to reach a search page (DuckDuckGo captcha) and an Aspden page (https://www.aspden.org/books/data/correa.htm, not found) failed; infinite-energy.com's search page returned unrelated content. So: no independent test or replication found in the sources I could reach, and the search was incomplete. This is "not found", not "none exists".
3. Lead-acid open-circuit voltage relaxation and surface charge [web: https://batteryuniversity.com/article/bu-903-how-to-measure-state-of-charge]. "Surface charge further fools SoC estimations by showing an elevated voltage immediately after charge; a brief discharge before measurement counteracts the error." For voltage-based state of charge "the battery needs to rest in the open circuit state for at least four hours; battery manufacturers recommend 24 hours for lead acid." The Correas' protocol allows a minimum of 15 minutes' relaxation, with no stated maximum (Carrell, relaying the patent; also read: p.17-19 of the printout via deep read 1: FIG. 13 shows voltage still relaxing). So the existence of a surface-charge and relaxation effect, and a rest time one to two orders of magnitude longer than used, is now [web]. What is still [memory, unverified] is that this effect is large enough to produce apparent gains of 4-34x (Table 8) or 218-563% (Table 7); a few percent of battery capacity is the size at stake, and no source I read gives the magnitude for these 12 V, 6 Ah gel cells. Direction: surface charge after charging inflates the charge pack's apparent state, and recovery after discharge lifts the drive pack's apparent state, so both errors increase apparent efficiency [memory, medium; the charge-side direction is [web]]. Note that Table 8's method maps discharge curves rather than reading open-circuit voltage, so the effect enters through the start of each discharge curve, which the authors' own FIG. 13 shows is unsettled.

Retagged claims. "Lead-acid relaxation and memory effects exist" [memory, medium] becomes [web] (batteryuniversity.com above). "Relaxation could account for the Table 8 gain" stays [memory, unverified]: it is the main ordinary explanation offered but no quantity has been shown. "No independent test exists" is replaced everywhere above by "no independent test found" with the check list above; "Never properly tested" is kept for claim 3 because none was found. The 2007 PDF is retagged from "could not be read" to [web] reprint of inventors' data.

What deep reads 3 and 4 and the motor read found

• Deep read 3 (pp.29-35). Vacuum processing, gauge types, Examples 1-6 and the plate-area data; findings in New evidence item 7 above (Penning gauge at seal-off, correcting QE; Paschen scope; area effect with edge-field and wall alternatives; one tube per sealed group; Vb depends on the supply; 4.8e5 eV not touched here). The authors' own oil-vapour concession [read: p.31] means the diffusion-pumped tubes are not clean vacuum. The polarity asymmetry (2 A against 200 mA peak) and the Alzak-versus-H34 factor of 9 in counts, 20-30 in rate, are unexplored clues pointing to surface and oxide effects [read: p.30-31, p.34].
• Deep read 4 (pp.36-43). (a) Every Table 13 row reproduces as V x I_DC x t_pulse, so "input energy per pulse" counts only the conduction time; at the worked example's 560 V, 1.6 A and 0.3 pps the per-period input would be about 2,987 J against 253.5 J output if 1.6 A were continuous, but the text does not say it is, and the Carrell oscillograms above suggest it is not for at least some runs. The 28-60x ratios sit at 0.2 pps, as the omission would predict, which is also what a bank-discharge picture predicts. (b) Reactor energy (8, 14, 9 J rows: 367.7, 261.5, 71.5 J) is consistent with 0.5 C V^2 on the 17.5 mF bank at Vp 90-205 V, and the "dynamic capacitance 0.2-80 mF" is back-calculated as 2E/Vp^2; it brackets the external bank, so it is the bank, not the 2 pF tube [inference]. (c) The window, integrand and instrument bandwidth are not stated; 8 of 14 reactor entries are ND. (d) The worked example reproduces except that 86.4 J / 1.3e15 gives 4.2e5 eV, not the printed 4.8e5 (4.8e5 corresponds to 100 J); the 86.4 J cannot be traced to Table 14 and is a net electrical figure, not ion energy; the crater volume comes from different runs than the 86.4 J. (e) The 245.2 dyn reading has no apparatus, calibration, error or controls; peak versus time-average is unstated, a factor of about 70 at 0.3 pps and 50 ms. m v / 40 ms reproduces 245.06 dyn, within 0.06%, a curiosity that I record without drawing a conclusion. (f) Table 15 row 4 is garbled (F printed 0.9e11 dyn but k i^2 = 9.4e11); Table 12 row 5 has a slip (11,200 V/m recomputed, 12,200 printed); Table 16's "work function" is electronegativity converted by X = 0.44 phi - 0.15, not measured (aluminium about 4.06-4.26 eV [memory, medium]).
• Motor patent read (US 5,416,391 Examples 1-5, pp.11-13). The text carries almost no numbers; rpm, torque and power are in FIGS. 7-13, which are absent from the printout. Example 3 torque is standstill (locked-rotor) torque, so it gives no mechanical power. The 2,800-3,000 rpm at 10 pps remark is from US 5,449,989 (Table 7 run 12 context, "electrical flywheel"), not from 5,416,391; at 10 pps a 2-pole synchronous motor would run 600 rpm [calc], so the figure needs about 5 cycles per pulse (a ring-down in the motor-capacitor circuit is a plausible hypothesis, not shown). Stated pulse rates: maximum 30 pps (diode), 43-45 pps (triode). The patent makes no efficiency claim for the motor circuit. facts.json should record that FIGS. 7-13 are absent and fix the attribution.

Verdicts changed by Round 2. None of the headline verdicts changes. Wording changes: (1) claim 1 sealed tubes stay Never properly tested, now with the corrected gauge statement, the Paschen-scope paragraph and the area-effect alternatives (item 7); (2) "100x vacuum arcs" stays Internally inconsistent, now cited to deep read 4 rather than inferred (threshold 12-16 A, k about 0.6-1.1 at 150 A); (3) "0.5 MeV ions" stays Refuted by calculation, strengthened (86.4 J untraceable and net electrical; 4.2e5 not 4.8e5 eV); (4) the 245.2 dyn force stays Never properly tested; (5) the per-pulse "3 to over 50 times" energies stay as stated in the Claim 3 section (not an independent measurement; the V x I x t booking, the 17.5 mF bank energy and unstated integration window prevent them being read as net energy; weakened-but-open on the inter-pulse input question given Carrell's oscillograms); (6) Table 8 stays Refuted by calculation as an energy balance (the 98 W floor argument rests on the authors' own currents and does not depend on lead-acid relaxation), and the lead-acid effect is added as the leading ordinary explanation for any residual apparent gain, with no magnitude established; (7) claim 3 general net gain stays Never properly tested, because no independent test was found (the 2007 PDF is a reprint), and claims 4, 5 and 6 are unchanged.

Corrected claim list updates. Claim 1: replace "tubes sealed at 1e-6 to 2e-6 Torr (not gauge-verified)" with "(Penning-gauge value on the pumping system at seal-off, not re-measured in the sealed tube)". Claim 2: "0.5 MeV" is also untraceable to Table 14 and is 4.2e5 eV on the authors' own inputs; the "100x" ratio flips to about 0.6-1.1 at the reported >150 A pulse current; 245.2 dyn has no described method and a factor-of-70 peak-versus-average ambiguity. Claim 3: the per-pulse energies use a conduction-time-only input and an unstated integration window; the 2007 PDF is a Carrell reprint relaying Correa oscillograms, not an independent test. Claim 5: the 2,800-3,000 rpm remark belongs to US 5,449,989, not 5,416,391.