RankingsPrimerArbiterPhysicistMathematicianContrarianQEClaimsSources

Correa: Quality Engineering (QE) report, Step 4b

Date: 2026-10-03. Inputs checked: _text/manifest.json, all 17 digests in digest/, facts.json, claims.md, physicist.md, mathematician.md, contrarian.md, arbiter.md, primer.md, example.py, and spot checks against the extracted text in _text/. There is no deep/ folder: this project was reviewed before Step 0/1a existed, so Step 1 sampled the documents and no deep reads were run.

Tags: [read] = extracted text (file and line or page given); [read: page image p.N] = page rendered by QE and viewed; [calc] = QE arithmetic.

Verdict: MINOR

None of the six headline verdicts flips. But three things must change. (1) One open question that the arbiter and primer leave unresolved can be settled from a figure in the folder: the circuit of FIG. 9 shows that the Table 7 "drive pack watts" is the power lost in the ballast resistor. (2) Claim 1's rebuttal rests on Paschen's law, so it covers only the gas-pressure tubes. It misses the sealed 10^-6 Torr tubes, and it misstates their pressure as "~10^-4 Torr". (3) Several headline numbers and data items in the digests are missing from facts, claims and primer, including the 6,750% Table 5 run. All of this can be fixed by updating facts, the arbiter (a QE revision section) and the primer. Steps 2-3 do not need to be redone.

1. Coverage

Manifest document Pages Digested? Notes
5502354_Direct_current_energized_pulse_g.pdf (US 5,502,354) 38, all OCR Yes, 3 digests (pp. 1-38) facts.json still says "image-only, text_available none, page 1 read". That was true when Step 1 ran. It is now out of date: Step 0 OCR'd all 38 pages.
Eng-Conv-Sys-US5449989.pdf (US 5,449,989) 36, all OCR Yes, 3 digests (pp. 1-36) Same staleness. The 13 drawing sheets (pp. 2-14) include the decisive circuit, FIG. 9 (p. 6). Only FIG. 12 (mathematician) and p. 1 (Step 1) were ever viewed.
Paul & Alexandra Correa _ Pulsed Abnormal Glow Discharges (PAGD).pdf 44, text layer Yes, 11 digests (pp. 1-44) This file bundles six documents: press release, Axelrad essay, S3-001 excerpt, US 5,416,391, US 5,449,989 and US 5,502,354. facts.json gives only 5,416,391 its own entry. It folds the press release, Axelrad essay and S3-001 into one entry and gives no entry to the bundled copy of 5,502,354. Under "If one PDF bundles several documents, list each", this is a minor defect.
  • Nothing was skipped, and every chunk has a digest. Coverage: pass.
  • facts.json gives "text layer sampled" as how_read for US 5,416,391, and says only the abstract was read. The text was extracted, so the phrase is not false, but the patent's five motor examples (digest PAGD.03, p. 12) were never used in facts. Claim 5 is peripheral, so this is acceptable. It should still be recorded honestly as "abstract only".

2. Shortcuts

Item Method used Acceptable?
Claim 3 (over-unity), Tables 5-8 No deep read. All three reviewers and the arbiter read the tables directly from the text layer and recomputed them. Acceptable in substance. The primary text was read, not a digest. One defect: nobody read the circuit drawings that the open "topology" question turned on (see 3a).
Claim 1 (Fowler-Nordheim, FN), area effect, sealed high-vacuum tubes (US 5,502,354 Examples 1-2 and 4-6; Tables 2-6, 12) Step 1 sampling plus targeted reviewer reads of Table 12 and the FN passage Defect. The high-vacuum examples (10^-6 and 2x10^-6 Torr) were not read, so the Paschen rebuttal was applied to all tubes (see 3b).
Claim 2 per-pulse energies (Tables 13-15) Reviewers read the worked example and Table 15 Mostly acceptable. How the per-pulse "input/reactor/output" energies in Table 14 were integrated is still only inferred ("evidently", contrarian). That is a derivation resting on inference: minor defect.
US 5,449,989 and 5,502,354 scans Page 1 only (Step 1); one drawing sheet (mathematician) Was acceptable while the scans had no text. Now that Step 0 has OCR'd them, the original scans should be used to check the rexresearch transcription. QE did this for Table 7 (section 4).
Arbiter replication search Not run (budget exhausted); relied on reviewers Acceptable: three reviewers searched and the limit is stated honestly.

3. Missed material

3a. The FIG. 9 topology settles the Table 7 question (most important)

The arbiter and primer both list "what is Y and how is the circuit wired?" as the open question that decides whether Table 7 shows a gain or a loss. QE rendered US 5,449,989 sheet 5 [read: page image p.6]. The drive pack DP connects through diode D1 and the ballast potentiometer R1 to the anode, and returns from the cathode through D4. The output network (C3, C5, the bridge D2/D3/D5/D6, C7a/C7b, D7/D8, and the charge pack CP at E1-E2) is coupled to the tube only through capacitors C3 and C5. So no DC path joins the two battery packs. The drive pack's mean power is X x I, where X is its terminal voltage and I its current. Y is "the DC voltage across the plates" [read: PAGD.txt l.1712]. X - Y is therefore the drop across R1 and D1.

The original scan's Table 7 confirms it [read: Eng-Conv-Sys OCR, p.34]. Column 9 "PAGD Volts" = X - Y (row 1: 562 - 350 = 212). Column 14 "Total Resistance" = (X - Y)/I (212/0.65 = 326 ohms printed; row 2: 160/0.60 = 267, printed 270) [calc]. That is the size of the ballast (the text quotes R1 = 600 ohms and a ballast range of about 150-800 ohms). So the printed "DP Watts" = (X - Y) x I is, to good approximation, the power dissipated in the ballast resistor, not the power the drive pack delivers. The arbiter's "natural reading" (input = X x I) is the correct one. On that basis 10 of 14 runs are at or below 100% (median about 92%, range 57-146%, as the mathematician and arbiter computed).

Consequence: for "Table 7 as printed", Refuted by calculation is now justified alongside Internally inconsistent: the stated input omits most of the drive pack's output, by the authors' own circuit. The four runs at 104-146% on terminal power stay unexplained. The likely candidates are rms meters on pulsed charge currents and battery effects. The general net-gain claim stays Never properly tested, because Table 8 does not depend on this column.

3b. Sealed high-vacuum tubes: the Paschen rebuttal does not reach them

  • US 5,502,354 Example 1 (device #1): 128 cm^2 plates, 5 cm apart, "10^-6 Torr at time of seal off". The tube showed PAGD from about 10 to 150 mA RMS with Vb plateauing at about 850 V [read: PAGD.txt l.1097-1102]. Example 5/10: sealed tubes at 2x10^-6 Torr. The 16 and 64 cm^2 tubes did not respond even at 3.3 kV, while a 256 cm^2 tube fired at Vb = 287 V [read: PAGD.txt l.1220-1223; digest 5502354.02 p.26].
  • At 10^-6 Torr, pd is about 5x10^-6 Torr cm. That is far to the left of the Paschen minimum, where gas avalanche breakdown at a few hundred volts cannot happen [memory, high]. So the reviewers' and primer's "tubes run at 0.09-1.5 Torr, so it is just gas breakdown" covers the energy-conversion runs (0.09-1.5 Torr, Tables 5-8) but not the high-vacuum emission claims. The arbiter and primer call the unanalysed case "sealed ~10^-4 Torr". The text says 10^-6 to 2x10^-6 Torr.
  • The mainstream answer for these tubes is different, and the primer does not give it: vacuum breakdown triggered by field enhancement, microparticles, desorbed gas and electrode outgassing. There is also a strong reason to doubt the pressure. Pressure was measured with an Edwards thermocouple gauge that reads only down to 10^-3 Torr [read: digest Eng-Conv.01 p.17]. Below that, vacuum was judged with a Tesla coil ("black vacuum") and x-ray onset [read: PAGD.txt l.3128; digest PAGD.08 p.30]. The pressure inside a sealed tube after it has run a discharge is unknown. The plate-area effect makes this point too: the authors themselves say it contradicts Paschen scaling (Tables 2-6, k_A regressions) [read: digest PAGD.09 p.32]. That is a separate empirical claim that claims.md does not list.
  • Fix: in the arbiter and primer, restrict the Paschen argument to the gas-pressure runs, correct "~10^-4" to "10^-6 to 2x10^-6 Torr (seal-off, not gauge-measured)", and keep Never properly tested for the high-vacuum strong form. Primer section 4 (row "Why electrons are emitted"), the evidence table row 1 and Answer 1(b) need this change.
  • Minor: the mathematician says the patent's beta of 10^6-10^7 "cannot be reproduced" from a field ratio. The text actually derives beta from the FN-plot intercepts (0.0005 and 0.00005), and the authors themselves say treating PAGD like a short-gap vacuum arc is "incorrect" [read: PAGD.txt l.4252-4254]. That criticism should be dropped or reworded.

3c. Other numbers and items not carried into facts, claims, arbiter or primer

  1. Table 5 run 5: 6,750% "breakeven" with the motor arm running [read: PAGD.txt l.2370; text l.1667-1669 "extremely large breakeven efficiency of PAGD run 5"]. This is the largest efficiency in the documents and is not mentioned anywhere downstream. The contrarian's "Table 5 gives up to 776%" is wrong: 776% is Table 6 run 1 [read: l.2522]. In Table 5, run 6 = 96% and run 8 = 906%.
  2. Cathode as "fuel", and energy per gram. The authors call the cathode erosion "a fuel consumption process" [read: digest PAGD.02 p.4; PAGD.txt l.410]. They give 0.25 kWh per 10^4 pulses and 20-40 MWh per H34 cathode [read: PAGD.txt l.4333]. QE: 86.4 J / 5.8x10^-8 g = 1.5x10^9 J/g [calc]. That is about 5x10^4 times the chemical energy of burning aluminium (about 31 kJ/g [memory, high]) and about 2% of uranium fission (about 8x10^10 J/g [memory, high]). It is a clean energy-budget check for the reader, and nobody uses it.
  3. The output battery sets the oscillation. Going from 29 to 31 cells in the charge pack cuts the PAGD frequency 10-fold, and adding more cells extinguishes it [read: digest Eng-Conv.01 p.20]. The PAGD rate also tracks the charge pack's state of charge (FIG. 5). This is important for interpreting claim 3, because the "output" is part of the oscillator. No review mentions it.
  4. AC rms vs DC current. At about 500 mA applied, the pulse AC rms current is up to 7.5 times the applied DC current [read: digest Eng-Conv.01 p.18]. The text also reports "current amplification" (>150 A peaks from 1.2-1.7 A input) [read: digest 5502354.03 p.34]. Together these strongly support the contrarian's open "rms meter" artefact. They should be cited in the arbiter.
  5. The authors' own statement that "the energy captured by this circuit at the output, is not directly related to that utilized in triggering the pulses from the input" [read: PAGD.txt l.1580]. Removing the C7 bridge suppresses PAGD altogether [l.1573]. Both are relevant to the topology discussion.
  6. People and events missing from facts.json: Ernesto Manuel's 1969 PAGD patent (US 3,471,316), A. Chernetskii's 1989 reactor (Novosti), Kimblin, Plyutto, Compton's <2% vapour-jet result, G. Spence's patent, Reich's "30 kc" claim, T. H. Moray, and the testimonials from W. Tiller, M. Carrell and U. Soudak. Also the second laboratory, ABRI (Aurora Biophysics Research Institute) [read: digest PAGD.01 pp.2-3]. These are peripheral, but the Curator step uses them.
  7. X-rays. At under 10^-6 Torr and higher voltages, PAGD gives way to "cathode rays and weak x-rays" [read: digest 5502354.02 p.23]. This is a safety point for a reader who might try to build one, and the primer's suggested experiment should mention it.
  8. Cathode lifetimes and Table 17 specifications (rates 0.01 to >50,000 pps, reaction energy up to 5x10^4 J) are not summarised anywhere. This is low priority.

4. Accuracy spot checks

# Fact checked Where used Result
1 Worked example: density 1.86 g/cm^3, 86.4 J "net" (= output reactor energy - input reactor energy), 3.1e-8 cm^3, 5.8e-8 g, 1.3e15 ions, 4.8e5 eV, 1.7e8 cm/s, F = m*V/1.39, 143.6 dyn, 245.2 dyn all reviews, arbiter, primer section 5, example.py Correct [read: PAGD.txt l.4300-4306]. Note the text says 86.4 J is "net" (output minus input), which strengthens the "circular" point.
2 Table 7 row 1 (X=562, Y=350, I=0.65, DP 137.8 W, CP 375 V x 1.25 A = 468.8 W, 340%); typos in rows 9 and 14 mathematician, contrarian, arbiter, primer Correct. The original scan shows the same values: row 14 prints 257.7 W, and row 9 has DP amps 0.75 / CP amps 0.65. So these are slips in the patent itself, not rexresearch transcription errors [read: Eng-Conv OCR p.34].
3 Table 8 range 406%-3,485%; group 6: 37.7 Wh loss, "182.2 kWh" gain, 483% contrarian, arbiter, primer Correct [read: PAGD.txt l.2676-2735]. The digest's "230% to 6,750%" for Table 8 (PAGD.07) is a digest error: it merges Tables 5 and 8.
4 Reactor pulse voltage Vp 100-500 V; tube capacitance about 2 pF; "dynamic capacitance 0.2-80 mF" physicist, arbiter (energy-per-charge check) Correct [read: PAGD.txt l.4285-4288].
5 "The sealed tubes are ~10^-4 Torr" and "the tubes run at 0.09-1.5 Torr" mathematician, arbiter, primer Incomplete/incorrect. The energy runs are 0.09-1.5 Torr, but the emission-claim tubes were sealed at 10^-6 and 2x10^-6 Torr (see 3b).

Digest reliability note. The OCR of the two scans drops superscripts ("4.8*10°eV" at 5502354 OCR l.3363). Haiku then guessed exponents in three places: "4.8x10^9 eV" (true value 4.8x10^5), "H34 lifetime 1.2x10^8" (true value 1.2x10^9), and the k ranges "10^9-10^12 i^2 vs 10^7 i^2" (the compilation gives 10^3-10^4 i^2 vs 10^2 i^2). No downstream file used these wrong values, but anyone reading only the digests of the scans should treat their exponents as unreliable.

5. Process

  • Source tags: present and mostly honest throughout. Reviewers mark [memory] with confidence and list what they did not web-check. The arbiter states openly that its search budget ran out.
  • Web checks: done for aluminium constants, cathode-spot ion energies (OSTI 810482), Paschen constants and replication searches. Defect: the lead-acid relaxation and "surface charge" effects are [memory, medium], and they are the main ordinary explanation offered for Table 8. The claim 3 verdict partly rests on them, so they need a web check. The contrarian found a third-party PDF (rohnermachine.com/Files/CPAGD.pdf, 2007) that nobody read. It should be checked before anyone says "no independent test exists". Mallove's Infinite Energy coverage of the Correas was also not searched for test reports.
  • Verdict scale: applied consistently, with one stretch. "Refuted by calculation (conditional on pulse current)" for the "100x vacuum arcs" claim depends on an unmeasured peak current at the time of the force. A conditional refutation fits the scale poorly. Better: "Internally inconsistent" (the force is normalised to the DC input current while the same text reports pulse currents 100 times larger), or give the conditional form plainly as a split verdict.
  • Primer vs STEP 4: all nine sections are present in order. Symbols are defined, prerequisites have textbook references, the worked example uses real numbers, the check-yourself answers are at the end, and Aspden's law is introduced with care. It fits the reader profile. Needed fixes: 3b (Paschen scope and the 10^-6 Torr tubes), 3a (resolve the open "topology" bullet in section 7 and the evidence-table row 3), and adding 3c items 1-3.
  • example.py: runs cleanly with .venv/bin/python (exit 0). Its outputs match primer section 5 (4.19e5 eV; 748x/249x/58x; k = 5,609 vs 0.6; 340% vs 128%).
  • facts.json: out of date on how_read/text_available for both scans (now OCR'd). The bundled documents are not listed separately. Key_claims for 5,449,989 cite only Table 7. They should add Tables 5, 6 and 8 and the 6,750% run.

6. Remediation actions

Deep reads (most important first)

  1. Eng-Conv-Sys-US5449989-60b35a, pp. 5-8 (Opus). Render FIG. 8, 9, 10A-C and 11 as page images. Confirm where X, Y and I were measured, and whether any DC path links the drive pack to the charge pack in the FIG. 10A-C variants used for Table 7. State whether "DP Watts" = (X - Y) x I is the ballast/D1 dissipation in every configuration.
  2. Paul-Alexandra-Correa-Pulsed-Abnormal-Glow-Discharges-PAGD-8687ea, pp. 17-19 (Opus). Tables 5, 6 and 8 and the four methods. Does Table 8 (pre/post resistive discharge) survive? Give relaxation times, the cutoff powers (115 W, 90 W), the cell counts, and how "Wh gain/loss" was integrated. Check Table 8 drive-pack losses against X x I x run time. Explain the 6,750% run 5 of Table 5.
  3. Paul-Alexandra-Correa-Pulsed-Abnormal-Glow-Discharges-PAGD-8687ea, pp. 29-35 (Sonnet). US 5,502,354 Examples 1-2 and 4-6 and the vacuum-processing description. At what pressures (and measured how) did PAGD occur in sealed tubes? What were the breakdown voltages and fields at 10^-6 to 2x10^-6 Torr? How do the area-effect data contradict Paschen scaling?
  4. Paul-Alexandra-Correa-Pulsed-Abnormal-Glow-Discharges-PAGD-8687ea, pp. 36-42 (Opus). Tables 12-16. How were the Table 14 input, reactor and output pulse energies computed (integration window, which capacitors, which currents)? What peak current coincided with the force readings? How was the 245.2 dyn "reaction weight" measurement made?
  5. Eng-Conv-Sys-US5449989-60b35a, pp. 30-35 (Sonnet). Check the rexresearch transcription of Tables 1-9 against the OCR'd original. List any differences that matter (row-level values, units, "kWh" vs "Wh").
  6. Paul-Alexandra-Correa-Pulsed-Abnormal-Glow-Discharges-PAGD-8687ea, pp. 11-13 (Sonnet). US 5,416,391 motor Examples 1-5. Report rpm against pulse rate, torque, any input/output power figures, and anything bearing on the 2,800-3,000 rpm at 10 pps anomaly.

Web checks (arbiter, at most 4)

  • rohnermachine.com/Files/CPAGD.pdf (2007): is it an independent test, a replication attempt or a reprint?
  • Any independent test or report in Infinite Energy (Mallove) or by Aspden on PAGD/XS NRG.
  • Lead-acid open-circuit voltage relaxation and surface charge after pulsed charging (retag [memory, medium] to [web]).

File changes

  • facts.json: update how_read/text_available for both scans ("OCR, all pages; digest"). Add entries for the bundled press release, Axelrad essay, S3-001 excerpt and the bundled copy of 5,502,354, or say clearly that one entry covers them. Add Tables 5, 6 and 8 (230-906%, 6,750% run 5; 776%; 406-3,485%) to 5,449,989's key_claims. Add the high-vacuum sealed-tube data and the plate-area/Paschen claim to 5,502,354. Change the 5,416,391 how_read to "abstract only (text available)". Add the people and events in 3c-6.
  • claims.md: in claim 3, give the full range by method (Table 5 to 6,750%, Table 7 218-563%, Table 8 406-3,485%). Add a claim or sub-claim: "Breakdown/sustaining voltage falls with plate area, contrary to Paschen scaling, and PAGD occurs in tubes sealed at 10^-6 Torr." Note that "0.5 MeV" is derived from a net pulse energy.
  • arbiter.md: add a "QE revision" section. (a) The FIG. 9 topology resolves the open question: Table 7 as printed becomes Internally inconsistent and Refuted by calculation (the input column is ballast dissipation), while net gain overall stays Never properly tested. (b) Correct "~10^-4 Torr" to 10^-6 to 2x10^-6 Torr (seal-off, gauge-unverified) and limit the Paschen argument to the gas-pressure runs. (c) Relabel or split "100x vacuum arcs". (d) Correct the contrarian's "Table 5 up to 776%". (e) Cite the charge-pack coupling and the AC-rms/DC ratio as support for the meter-artefact explanation. (f) Record the web-check results.
  • primer.md: section 4 rows "Why electrons are emitted" and "Energy balance"; section 5 Step 10 (note that DP watts is the ballast loss, with FIG. 9 described in words); section 6 rows 1 and 3; section 7 "Genuinely open" (remove the topology bullet, add the high-vacuum question); Answer 1(b) (limit to the gas-pressure runs). Optionally add the energy-per-gram check (1.5x10^9 J/g vs about 3x10^4 J/g for burning aluminium) to section 5 and example.py. Add an x-ray safety note to the suggested experiment.
  • example.py: optional Part D, the energy per gram of cathode and the Table 7 "Total Resistance" = (X - Y)/I check.

Recheck

Date: 2026-10-03. Checked arbiter.md (QE revision section), primer.md, claims.md, facts.json, example.py, deep/.

Defects

Item Status
3a FIG. 9 topology / Table 7 "DP Watts" Fixed. Deep read written; arbiter QE revision gives Table 7 as "Refuted by calculation and Internally inconsistent"; primer section 5, 6, 7 updated; example.py reproduces (X-Y)/I and 57-146%, median 92%.
3b Paschen scope, "10^-4 Torr" misstatement Fixed in text. No "10^-4" remains; primer and arbiter restrict Paschen to gas-pressure runs and give 1-2 x 10^-6 Torr (seal-off, not gauge-measured). Underlying deep read 3 not run (see below).
3b mathematician's beta criticism Not verified as reworded (minor); not counted.
3c-1 Table 5 run 5, 6,750% Fixed (arbiter, claims, primer, deep read).
3c-2 energy per gram Fixed (primer; example.py D1 = 1.5e9 J/g).
3c-3, 3c-4, 3c-5 pack coupling, AC rms/DC Fixed in arbiter QE revision.
3c-6 people and events Fixed (arbiter item 9, primer further reading).
3c-7 x-ray safety note in suggested experiment REMAINS. No x-ray mention in primer.md.
facts.json staleness, bundled documents split, 5,416,391 how_read Fixed (10 entries, container plus bundled docs, OCR status, "no deep read").
claims.md plate-area/Paschen claim and Table 5 range Fixed (claim 2 added).
Verdict-scale stretch ("100x vacuum arcs") Fixed: now "Internally inconsistent".
Contrarian's "Table 5 up to 776%" Corrected in arbiter revision.
example.py runs Fixed/OK: exit 0, outputs match primer.

Remediation actions

Action Status
Deep read 1 (FIG. 8-11, Table 7 columns), Opus Done (deep/Eng-Conv-Sys-US5449989-60b35a.md)
Deep read 2 (Tables 5, 6, 8, four methods), Opus Done (deep/Paul-Alexandra-...-8687ea.md)
Deep read 3 (US 5,502,354 high-vacuum examples, pp.29-35), Sonnet REMAINS: not run (arbiter admits it)
Deep read 4 (Tables 12-16, per-pulse energies, 245.2 dyn), Opus REMAINS: not run
Deep read 5 (transcription check Tables 1-9) Partly done inside deep read 1 (Table 7 only); rest not run. Counted as covered, minor.
Deep read 6 (US 5,416,391 motor examples), Sonnet REMAINS: not run
Web checks (rohnermachine PDF, Infinite Energy/Aspden tests, lead-acid relaxation) REMAINS: arbiter states none were made; facts stay [memory, medium] and the verdict on Table 8 / claim 3 still partly leans on them
File changes (facts, claims, arbiter QE revision, primer, example.py) Done, except the x-ray note above

Open items: (1) deep read 3, (2) deep read 4, (3) deep read 6, (4) web checks, (5) x-ray note in primer. No headline verdict is in doubt; the open items are completeness and sourcing.

Final: OPEN (5 issues)

Recheck 2

Date: 2026-10-03. Checked facts.json, claims.md, arbiter.md (QE revision, Round 2), primer.md, deep/ (two files; the second QE round sections added).

Remaining item from Recheck Status
Deep read 3 (US 5,502,354 Examples 1-6, vacuum processing, pp.29-35) Fixed. Section "QE round 2: US 5,502,354 Examples 1-6 and vacuum processing" now in deep/Paul-Alexandra-...-8687ea.md; primer section 4 row 1 uses it (1-2 x 10^-6 Torr, gauge-unverified, ionisation count).
Deep read 4 (Tables 12-16, per-pulse energies, 245.2 dyn) Fixed. "Second QE round (deep read 4, Opus)" in the same deep file: window and integrand unstated, 245.2 dyn has no apparatus or error, 86.4 J cannot be traced.
Deep read 6 (US 5,416,391 motor Examples 1-5) Fixed. "QE round 2: US 5,416,391 motor Examples 1-5" in the deep file; the 2,800-3,000 rpm figure is shown to come from US 5,449,989, not this patent.
Web checks (rohnermachine PDF, Infinite Energy/Aspden, lead-acid relaxation) Fixed to the extent possible. Arbiter "Round 2 web checks" reads the 2007 PDF (a reprint of Carrell's Infinite Energy articles, not a test), cites batteryuniversity.com for surface charge/relaxation (retagged [web]), and states plainly that the Infinite Energy/Aspden search was incomplete ("not found", not "none exists"). Claim 3 stays "Never properly tested".
x-ray note in primer suggested experiment Fixed. primer.md now has a safety paragraph (section 5) and a safety note in section 7 citing the x-ray and cathode-ray reports and ruling out sealed high-vacuum tubes without shielding.

Remaining after Recheck 2: none that changes a verdict. The incomplete Infinite Energy/Aspden search is disclosed in the arbiter as a limit, not a defect. Deep read 5 (full transcription check of Tables 1-9) was already counted as covered in Recheck.

Final: PASS