Arbiter: Ainslie-Rosemary-Ann

Inputs: facts.json, claims.md, physicist.md, mathematician.md, contrarian.md. I re-read the 2009 reprint text layer for the points where reviewers differed [read: pdftotext electric_heater_experiment.pdf].

Source gap (revised after QE; see "QE revision" below). The raw oscilloscope data, the ABB / Cape Hope Metrology Laboratory / BP test reports and any neutral replication are not in the folder. The patent bundle (43 pages), the Quantum article and the Pretoria News clipping now have OCR text; the patent Examples 1-3 and the Quantum article were deep-read after QE. Conclusions about the "independent tests" and about the Stellenbosch refusal still rest on the author's own report, a third-party digest, or [memory]. (The original text of this paragraph, written before the OCR existed, said the patent could only be read through Kelly's digest; that no longer applies.) Conclusions about the experiment itself rest on the authors' published tables plus my arithmetic.

Reconciling the numbers

All three reviewers agree on the arithmetic that matters; two points needed recomputation.

1. Energy total. Reported average load power 17.53 W x 59,820 s (997 min) = 1.049 MJ, not the 1.22 MJ in the paper. 1.22 MJ is 20.44 W x 59,820 s, i.e. the last table row's instantaneous watts, and the table's "Joules" column is (instantaneous watts x elapsed time), not an integral (checked by physicist and contrarian; the mathematician's trapezoid check agrees through 960 min) [read; arithmetic mine]. Corrected ratio: 1.049 MJ / 67.6 kJ = 15.5. Adding the 12 V gate-supply battery (12.45 V x 0.039 A = 0.486 W, x 59,820 s = 29.1 kJ) gives 1.049 MJ / 96.7 kJ = 10.8 [arithmetic mine, run in the venv]. The 0.039 A is described only as "consistent with" the gate battery's voltage fall (12.45 V to 12.04 V open-terminal); the paper gives no measurement method for it (QE revision; replaces the earlier wording "inferred") [read: Quantum journal p.8 = PDF p.5].
2. What ordinary theory predicts: a reviewer error. The physicist and mathematician computed battery power at 1.3% duty as 0.73-0.75 W by assuming the current reaches its steady value 2.34 A. That is wrong at 143-200 kHz: the on-time is only about 76 ns, far shorter than L/R = 8.64 uH / 10.25 ohm = 0.84 us, so the current only reaches about 0.2 A (contrarian's figure, confirmed). I simulated the periodic steady state (ideal diode freewheel, L = 8.64 uH, R = 10 ohm + 0.25 ohm shunt, 24 V):

Reading: no stated operating point gives "1.13 W in, 17.5 W heat". The stated 3.7% setting predicts about 2 W in and 2 W of heat (the reported 1.13 W is within a factor of two of that); the 55% duty Kelly describes predicts about 18 W in and about 18 W of heat, which matches the measured heat but not the measured input. The simplest reading is that the heat measurement reflects a high effective duty (about 30%, battery-fed) and the battery-power measurement is wrong or covers a different state. That is an inference, not a demonstrated cause; the raw waveforms are not available. 3. Maximum inductor energy. (1/2) L I^2 = 23.7 uJ at I = 24/10.25 = 2.34 A (physicist; mathematician's 25 uJ uses 2.4 A). At 200 kHz an upper bound of 4.7 W if the full energy were dumped every cycle, and it was drawn from the battery first. Agreed by all three. 4. Gate drive check. Qg x f = 190 nC x 200 kHz = 0.038 A, matching the reported 0.039 A, so the 143-200 kHz figure is plausible [web: https://www.vishay.com/docs/91254/91254.pdf and search summary giving Rds(on) 2.0 ohm at 10 V, Qg max 190 nC; the datasheet itself I did not open, search summary only]. Rds(on) of about 2 ohm raises the effective series resistance (the physicist's 32% effective duty becomes about 38%); no verdict depends on this. 5. Battery voltage is non-discriminating (all three). 480 Wh nominal = 1.73 MJ; 1.05 MJ would be about 60% of capacity (about 12 Ah), which is detectable, but the authors omitted the ampere-hour check, calling battery evaluation subject to "too many vagaries" [read: 2009 reprint lines ~473-475]. So "no measurable voltage fall" in claims.md is too strong; the paper says no "significant" reduction.

Reconciling reviewer disagreements on sourcing

• Who says the energy comes from the environment (claim 4). Mathematician is correct and physicist/claims.md were loose: the 2009 reprint says "nowhere is there a claim to have a closed system" and the appendix says energy delivered by the battery goes back to the battery and load energy "comes from the load" (bound magnetic fields in the resistor) [read: reprint lines ~618, ~795]. The "environmental energy" reading is Kelly's [read: Ainslie_heater_circuit.pdf per mathematician]. Ainslie's own reading is itself nowhere derived mathematically.
• Company support (claim 6). All three agree: Cape Hope calibrated instruments; Spescom attended demonstrations; ABB tests "confined to" instantaneous power with no result stated; BP applied tests on battery duration with no result stated; Sasol offered a bursary [read: Quantum p.8, Pretoria News]. These are not confirmations. OpEdNews (an advocacy piece) says "five commercial laboratories fully verified the claims" without naming results [web: https://www.opednews.com/articles/The-Strange-Case-of-the-Ro-by-Steve-Windisch-ji-091219-425.html]; I treat that as unsupported.
• Replications (claim 1). The contrarian found an advocacy article reporting that "at least three" people reproduced the heating effect in 2009, including Glen Lettenmaier [web: same OpEdNews URL; fetch summary did not give COP figures. The contrarian's "COP greater than 4" could not be confirmed by my fetch, so I do not use it]. On the other side, a forum analyst (Poynt99) traced the circuit and reported measurement artifacts, including a discrepancy between the probe-ground position in the publication and in the build [web: search summary of https://overunity.com/13743/rosemary-ainslie-quantum-magazine-circuit-cop-17-claims/30/ and https://www.youtube.com/playlist?list=PLml9VdOeqKa_5QWbelpO5Wp8MctLt5EdQ; I did not open either, so low reliance]. Neither side is a neutral, controlled, published test. No verdict depends on a replication having happened or failed.
• Verdict labelling for claim 1. Physicist: Refuted by calculation (energy source) + Never properly tested (measurement). Mathematician: Internally inconsistent + Never properly tested; declines "Refuted" because the raw waveform is missing. Contrarian: Never properly tested; strong form Refuted by calculation. I side with the split below. The scale says "Refuted by calculation" applies when accepted laws applied correctly rule out a claim; conservation of energy for a circuit whose only sources are a 24 V battery and a 12 V gate battery does that for the claim of a net energy gain. The mathematician's caution is right about something different: the measurement as an instrument reading (what the scope showed) cannot be called wrong without raw data, so that part gets "Never properly tested", and the published energy column is separately "Internally inconsistent".

Claim-by-claim

Claim 1. About 17.5 W of load heat from 1.13 W of battery power (gain 16-18), over 16.5 h, no battery voltage fall

• Agree: V^2/R = 13.32^2/10 = 17.74 W and 1.13 W x 59,820 s = 67.6 kJ are right; the 1.22 MJ is not; load power is a calibrated thermal measurement (sound in principle; "crude", 10% allowed, and the linear 0.34 W/K law is assumed across 16-60 K) [read]; input power is a scope shunt measurement over a 1.2 microsecond sample window from a 0.25 ohm shunt (47 mA mean is about 12 mV: contrarian), single-ended, with sign-split netting that depends on the zero level (mathematician C-E).
• Disagree: label only (above).
• Stronger argument: the mathematician's and contrarian's shunt-fragility case, strengthened by my simulation, which shows heating of this size needs about 30-55% effective duty and then the battery must supply about the same power. The physicist's "must have a sample-window error" is plausible but unproven.
• Unexplained: load temperature rises from 48.7 C to 59.9 C in the final 37 minutes with steady battery voltage [read: physicist]; the authors do not comment.
• Verdict. (a) Net energy gain (heat out exceeding battery plus gate-supply energy): Refuted by calculation (energy conservation; stated components; the energy stored in L cannot exceed what the supply put in). (b) The reported ratio as a measurement: Never properly tested by a neutral party (authors' tests only; not dismissed untested: examined by the authors, unnamed companies and forum builders, but no controlled neutral test with an ampere-hour count); corrected ratio is 15.5 on the author's own battery-side numbers and about 10.8 with the gate supply. (c) The paper's published energy figure (1.22 MJ, "greater than 16"): Internally inconsistent with its own Watts column.
• Tested / never tested / dismissed: tested only by the authors and unverifiable company reports; the cheap decisive test (ampere-hours, calorimeter plus coulomb counter) was never run.

Claim 2. The gain comes from back EMF, returning energy partly to the battery, growing with frequency

• Agree (all three): back EMF (V = L dI/dt) and flyback recirculation are Established; as a net energy source it fails the energy bound; the frequency dependence is asserted with no data, and the appendix itself says no part of the model is developed mathematically [read].
• Note: D1 is across the load (physicist), so current recirculates locally; return to the battery occurs only via the MOSFET body path.
• Verdict. Back EMF existing and recycling energy: Established. Back EMF as a source of net energy: Refuted by calculation. Frequency scaling of gain: Never properly tested (and my simulation shows that frequency matters only through the effective duty and edge effects, an ordinary mechanism).
• Never tested vs dismissed: never tested (no frequency sweep table in the paper).
• Added in QE revision: the patent's own bench tests (Examples 1-3) are claim 2's earliest experimental support; see "QE revision" below.

Claim 3. Tuned gate resistor causes self-oscillation far above the 555 setting, and that produces the gain

• Agree: parasitic oscillation of a MOSFET with a gate resistor and drain inductance is ordinary engineering [memory, high]; the gate-drive current check (0.038 A vs 0.039 A) supports about 200 kHz. The record gives three states: 2.4 kHz / 3.7% (2009, plus "random oscillation"), 143-200 kHz / 1.3% (2002), over 500 kHz / 55% (Kelly). The mathematician refined the physicist's and Step 1's reading: the 2009 paper does not merely repeat 2.4 kHz, it says adjusting the gate resistor gives "random oscillation that overrides the predetermined frequency and duty cycle" [read].
• Disagree: whether the claim is "Internally inconsistent" (physicist) or merely "Not testable as stated" (mathematician). The three states are described as one experiment, so the record is inconsistent as a record; the causal claim cannot be tested until the state is fixed. My simulation shows a 55% / 500 kHz state would heat the load at about 18 W while drawing about 18 W, which means changed effective duty alone explains the heating.
• Verdict. Oscillation itself: Plausible, untested (no neutral spectrum on record). That it produces extra energy: Never properly tested (and mainstream explanation, higher effective duty plus input under-measurement, is sufficient). The documented description across sources: Internally inconsistent as a record of a single operating point.

Claim 4. Unity is breached; the second law does not constrain EM energy transfer; extra energy from an unseen medium

• Agree: no mathematics; "magnetic fields comprise particles, so the second law does not apply" is a non sequitur; first-law energy conservation and Poynting's theorem are untouched by any evidence cited. The paper says it hopes the absence of references "will obviate the need for references" [read].
• Resolved: the environmental-energy source is Kelly's, not Ainslie's; Ainslie disclaims a closed system, and the reprint's own claim is that the resistor's bound fields supply the heat.
• Verdict. Unity breached by this circuit: Refuted by calculation (see claim 1). Second law not constraining EM transfer (as argued): Internally inconsistent. Environmental/unseen-medium or bound-field energy source: Not testable as stated (no measurable prediction), although a shielded or calorimetric closure test could be designed.
• Never tested vs dismissed: nothing specific to this was tested.

Claim 5. Current is moving magnetic dipole particles ("zipons", tachyons at about 2c); proton = three electrons; model predicted over-unity and unifies non-locality, dark matter, mass ratio

• Agree: no mathematics by the author's own statement; no measurable prediction apart from the heater result; deep inelastic scattering shows quark substructure and charge conservation forbids a proton made of three electrons (charge -3e vs +e) [memory: high; not re-verified on the web, standard textbook physics].
• Mathematician's added detail (uncontested; sharpened in QE revision): the mass-ratio sum is 108 + 1728 = 1836 with hand-picked factors, and the author's text writes "1837.5, minus 1.5 ... 1836", an add-then-subtract of 1.5 that was never listed as a summand; with the electron at 1.5 units a 1836-unit proton gives 1224, not 1836; measured ratio 1836.15 [memory: high].
• Disagree: label for "proton = three electrons": physicist "Tested, not supported" (scattering); contrarian "Refuted by calculation"; mathematician "Internally inconsistent". Each is defensible. Charge accounting is a calculation using accepted laws, so I use Refuted by calculation for the charge point and Internally inconsistent for the author's own mass-ratio arithmetic; I do not use "Tested, not supported", because no experiment was designed to test the author's specific model.
• Verdict. Claim that the heater result supports/predicted the model: Refuted by calculation (the supporting experiment fails the energy budget) and the paper itself says it does not conclusively prove it. Proton = three electrons: Refuted by calculation (charge, spin, accepted scattering results). Mass-ratio derivation: Internally inconsistent (units give 1224). Zipons, tachyonic dipoles, resolving non-locality and dark matter: Not testable as stated.
• Never tested vs dismissed: never tested; not dismissed in any documented way.

Claim 6. Institutions refused to test (Stellenbosch refused a demonstration) while companies supported it

• Agree (all three): the Stellenbosch refusal is a self-report in the Pretoria News (14 Nov 2002) with no university reply; the clipping also notes Ainslie "couldn't give us the oscilloscope" for the exhibit; named companies only "approved the inclusion of their name" and recommended wider, systematic consideration [read: Quantum p.8, page image Pretoria News]. I found no independent web record of the Stellenbosch refusal or the ABB/BP reports (checks failed or only advocacy: https://www.opednews.com/...425.html repeats "refusal of South Africa academics" without a source).
• Contrarian's distinction kept: refusal to look (self-reported, unverified, possibly dismissed without testing) is separate from tested and failed (not shown) and from never properly tested (the applicable class for the experiment).
• Verdict. Neutral testing: Never properly tested; whether Stellenbosch dismissed it without testing is unverified (one-sided source). Corporate "support": not supported by the record (named mentions, no reports). Not scorable on the physics scale beyond this.

Corrected claim list

1. The circuit, as reported by its authors, showed about 17.5 W of load heat (calibrated thermally, 10% stated error, crude conditions) against a scope-derived battery input of 1.13 W; the authors' own energy figure should read about 1.05 MJ, not 1.22 MJ, giving a ratio of about 15.5, or about 10.8 when the 0.49 W gate battery is included. The result has not been neutrally replicated; the gate-battery current has no stated measurement method; energy conservation says the input measurement is wrong (verdict: net gain Refuted by calculation; reading Never properly tested; published energy figure Internally inconsistent).
2. Back EMF and diode recirculation exist and return energy; they do not produce net energy, and no frequency data were given. The patent's three bench tests (COP 1.086; 2.45 or "100 percent above unity"; load/battery 0.976-1.176) use three different methods, none shows a net gain, and each has a stated defect (Example 1: 8.6% inside reading error; Example 2: averaged-as-rms slip and a discarded battery-2 trace that breaks Kirchhoff's current law; Example 3: table inconsistent with its own 6.3 ohm load).
3. Detuning the gate resistor produces parasitic oscillation, described in the record as three different states (2.4 kHz / 3.7%, 143-200 kHz / 1.3%, over 500 kHz / 55%); changed effective duty explains the extra heating without extra energy.
4. Ainslie claims the result breaches unity and is not constrained by the second law because magnetic fields are particles; she disclaims a closed system. The "unseen environment" energy source is Kelly's reading. The second-law argument is a non sequitur and the measurement fails the first-law budget.
5. The magnetic-dipole "zipon" model is verbal and unquantified; its proton-mass sum is hand-fitted, adds then subtracts 1.5 (1837.5 - 1.5 = 1836) and fails its own units; proton = three electrons contradicts charge conservation.
6. Ainslie reports that Stellenbosch refused a demonstration and that companies attended, calibrated or gave limited tests; no neutral test or company test result is on record in the folder, and the companies did not endorse the 17x figure.

What would settle it

• One experiment (decisive for 1-4). Run the same circuit from a battery with a coulomb counter (ampere-hours) or from a charged capacitor bank (E = 1/2 C (V1^2 - V2^2)), include the 12 V gate supply, and measure load heat calorimetrically against the DC calibration; or log full waveforms with a differential probe across a low-inductance coaxial shunt over an integer number of cycles. If heat exceeds total energy drawn, that is a first-class anomaly; if not, the artefact explanation stands. All three reviewers independently proposed variants of this.
• One calculation. Re-derive battery and load power from the authors' raw waveform samples with the positive-part-only and total (signed) integrals and the true cycle count; compare to the effective-duty table above. Raw data are not in the folder.
• For claim 5: any single quantitative, falsifiable prediction from the model other than the heater result.
• For claim 6: the ABB, BP and Cape Hope reports, and the Stellenbosch correspondence.

Open questions

• What was the actual operating state (frequency, duty, gate waveform) during the 16.5 h run, and was it stable?
• Why did load temperature rise by about 11 K in the last 37 minutes at steady battery voltage?
• What do the ABB and BP reports actually say, and what was the battery-duration result BP measured?
• What was the sign and size of the offset on the shunt signal, and where was the probe ground in the build versus the published figure?
• Is the 3-person 2009 replication claim (advocacy source) backed by data, and did any replicator include a coulomb count or calorimeter?
• Was the Stellenbosch refusal a refusal to test or a refusal to host a demonstration?

Review-process notes

• Reviewer error corrected: physicist and mathematician's 0.73-0.75 W at 1.3% duty assumed steady-state current; at 76 ns on-time the current reaches only about 0.2 A (contrarian correct), giving about 0.03-0.04 W. This changes no verdict but removes an apparent consistency between the 1.3% mode and the 1.13 W reading.
• Not verified on the web: the contrarian's "COP greater than 4" attribution to Lettenmaier; Poynt99's detailed findings (search summaries only); [memory] facts on deep inelastic scattering and the lack of observed tachyons (standard physics, high confidence, not retagged).
• Web calls used: 4 of 8 (none added in the QE revision; its new evidence is [read] from the deep reads plus my [calc] arithmetic).

QE revision

Trigger: qe.md verdict MINOR. New inputs: deep/harnessing-a-back-emf-...md (patent pp.19-32, Examples 1-3), deep/quantum-october-2002-...md, deep/magnetic-field-model-...md (pp.17-22). No headline verdict changes. Tags: [read] = deep reads of the document; [calc] = deep reader's or my arithmetic. No web checks were needed; no verdict rests on a [memory] fact.

New evidence: the patent's own bench tests (claim 2, earliest experimental support)

Dates: priority ZA 98/0541 filed 22 January 1998; bundle dated 19 January 1999; WO 99/38247 per Kelly [read: patent p.1, per QE]. Call it "1998 (ZA) / 1999 (WO)". Its claimed regime is 50-99% duty (preferably 75-95%) at 40 Hz or more, the opposite of the 1.3-3.7% heater circuit [read, per QE]. Patent pages: printed page = PDF page minus 2.

1. Example 1 (pp.19-22). Control: 68 ohm in series with a 12.28 V battery, 95 C, battery power 2.406 W (0.196 A x 12.28 V). Test: 22 ohm resistor in parallel with an inductor, switched by a MOSFET at about 5 kHz (0.2 ms period), no diode; duty adjusted until 93 C; mean battery current 0.1718 A, 2.1097 W. A 5% allowance gives 2.215 W and COP 1.086 (without it, 1.14). The method is consistent (battery V x average current on both sides) but has three defects [read; calc]: an arithmetic slip ("0.043 V" should be 0.0043 V; the final current is right); the 8.6% is no bigger than the scope-reading error on a 5-19 mV sloped waveform (the 5% margin is applied to one side only); and heat is judged by matching a single temperature, with no calorimetry (the patent admits this on p.29) and an unstated ambient. By conservation the load heat cannot exceed battery energy minus inductor, MOSFET and wire losses, so the waveform is ordinary parallel R-L switching.
2. Example 2 (pp.22-29). Control: 100 ohm on 6.12 V (two 6 V batteries in parallel), 0.0612 A, 0.374 W, 30 C (ambient 22 C). Test: 47 ohm resistor in series with an inductor, diode and MOSFET, 50% duty, same 30 C. Three sense resistors of 0.05 ohm: SR1 (load), SR2 (battery 1), SR3 (battery 2). Defects [read; calc]: (i) the load "rms" is peak/2 = 0.003 V, which is the mean of a triangle, not its rms (peak/sqrt 3); (ii) the claim that battery V x load I_avg is the load power is a category error: battery V x I_avg is right for the source only; (iii) the SR3 trace (+-5.5 mV = +-0.11 A, the largest current signal in the circuit) is "discounted" as "too small"; with SR2 alone the battery power is 0.153 W (0.025 A x 6.12 V), but the three waveforms do not satisfy Kirchhoff's current law (the buck relation expects about 0.030 A mean battery current, 0.184 W), and the discarded SR3 mean decides whether COP is above or below 1; (iv) two incompatible results appear: about 10% gain (0.169 W / 0.153 W) and "100 percent greater than unity" (0.0612 A / 0.025 A, which is actually 2.45); (v) the 47 ohm part reaching the same rise as the 100 ohm part at about 0.17-0.23 W electrical points to non-resistor heating or a different thermal resistance, not extra energy. QE's description of this example as simply "dropping a battery" is correct in effect; the deep read adds the KCL point.
3. Example 3 (pp.30-32). 12 V battery, 0.05 ohm sense resistor S1, 6.3 ohm 50 W load R1, 350 VA coupled inductor with a diode through the second winding, duty 90% to 50%. Battery power = V x average current through S1; load power = V3-4 rms x I rms. Correction to qe.md item 3: this is NOT an average-versus-RMS mix. Using net average current for the source and rms x rms for a resistor is the correct method. The defect is in the measurements: V3-4 rms / I rms (which must equal R1 = 6.3 ohm) is 4.89, 5.18, 5.24, 4.68 and 2.96 ohm at 90, 80, 70, 60, 50% duty. With R1 = 6.3 ohm (V^2/R) the ratios fall to 0.91, 0.94, 0.86, 0.80, 0.46, all below 1, as energy conservation requires. At 50% duty the net battery current is 1.2 mV across S1 (0.024 A) against 15.9 mV rms (I rms / I av = 13), so a 1 mV scope offset changes battery power by 100%. The table's printed ratios are 1.176, 1.148, 1.033, 1.082 and 0.976 (the text says "above unity for duty cycles above 60%" but 60% is itself above 1; the OCR "4.022" in the 70% row is 1.022 on the page image) [read: page image p.32; calc].
4. The three examples do not corroborate one another. Different supplies (12.28 V; 6.12 V; 12.57-12.76 V), different circuits, methods, instruments and efficiency figures (8.6%; 10% or 145%; 0.976-1.176). Only Example 1 gives a frequency. All say the values are "not necessarily optimized".
5. Verdicts on the patent examples. Example 1: Never properly tested (a single-point temperature match; the 8.6% sits inside reading error; not dismissed, but not above-unity evidence). Example 2: Internally inconsistent (rms slip, incompatible efficiencies, KCL violation); the claim that V x I_avg is load power: Refuted by calculation (I^2 R, i.e. rms, is the definition of resistive power). Example 3: Internally inconsistent (the table contradicts its own resistor); with the correct R1 the data show no gain. Net gain from any of the three: Refuted by calculation by the same energy bound as claim 1, with the caveat that the raw waveforms are not available. A coulomb counter (ampere-hour count) on the battery would settle it, as for the main experiment.

Other points from the QE and the deep reads

• Sample window and count [read: Quantum journal p.6-7]: the 1.2 microsecond window is stated; the oscilloscope memory is "about 1 200 simultaneous real time samples", a different quantity. No sample count for the 1.13 W figure is given. At 143-200 kHz, 1.2 us is about 17-24% of one cycle.
• Gate-supply current: "inferred" is dropped in favour of "method not stated" (spot check 3, deep read of Quantum). Roughly 28 kJ (about 42% on top of 67.6 kJ) if 0.039 A is right; ratio about 11 on the 1.05 MJ figure, as already given above.
• Claim 3 wording: the Quantum paper says results depend on "the precise components used" (journal p.5) and then that they are "easily repeatable" (p.8) [read].
• Claim 5 arithmetic: the author's text reads "6 x 6 x 3 = 108 ... 24 x 24 x 3 = 1728. This gives a mass of 1837.5, minus 1.5 for the quarks ... giving a total of 1836" [read: magnetic field model p.19]. 108 + 1728 = 1836, so the 1.5 is added unseen and removed; the target 1836 is built in by the free factor 4 (factor 3 gives 607.5; factor 5 gives 4387.5) [calc]. The quarks are given mass 0.5 each and then called massless. Charge accounting is also absent: three electrons lose one and a proton (+e) results [read: pp.18-19]. Verdicts for claim 5 unchanged; the "Internally inconsistent" label gains the 1.5 add-subtract and the unexplained factor 4.
• Page citations: journal page = PDF page + 3 for the Quantum article (PDF p.1 is the cover advert; folios on PDF pp.2 and 4 seen directly, pp.3 and 5 inferred). Earlier "Quantum p.7 / p.8" citations in this file and in the reviews are journal pages; "p.5" above means the same.
• Kelly's other claims (prototype above 4x input, "COP 4 to 17", self-charging lamp/battery variant): third-party advocacy with no data in the folder; no verdict beyond "Never properly tested". Donovan Martin (2009 co-author) and Brian Buckley (2002 co-author) are the named collaborators.
• 2009 appendix: says the load energy "comes from the load" (bound fields) and that delivered energy returns to the battery; already handled under claim 4.

Changed verdicts (table)