QE (Step 4b): Larson-Tyson (Realm Industries)

Reviewer: QE (Opus). Date: 2026-10-04. Inputs read: _text/manifest.json and _textprep.log, all 6 digests, both deep reads, facts.json, claims.md, sites.md, all 7 research notes (research/files/ is empty), physicist.md, mathematician.md, contrarian.md, arbiter.md, primer.md and example.py. Spot checks were made against the extracted text in _text/ and against two page images rendered to /tmp/qe-larson: p.1 of US 2009/0092540 and p.13 (Fig. 12) of US 2009/0090312.

Verdict: MINOR

The central verdicts hold. None of the defects below would change a final verdict in arbiter.md. They are attribution, retention and wording problems, fixable by updating facts.json, claims.md, arbiter.md and primer.md. One of them is an attribution error in the only independent document in the folder, carried forward from the digest. It does not move a verdict, but it changes who said what, so it must be fixed.

---

1. Coverage

No document was skipped, and no browser junk is present. research/files/ is empty: the Investigator says no downloads were made in this run, and the PDF links are recorded in the notes. Coverage: pass.

2. Shortcuts

• Every central claim rests on a deep read, not on a digest alone. Claims 1-3 rest on the full deep read of 0092540. Claim 4 rests on deep reads of both patents' Figs. 11-13 from page images, and the contrarian also re-rendered p.12 independently. Claim 5 rests on the verbatim examples. Claim 6 rests on the article.
• The 0092540 digest names the second inventor as "John Larson". The deep reader corrected this from the page image, and the QE confirmed it: p.1 reads "William Stehl ... Tyson Larson". This is the expected digest error, caught as the process intends. Nothing downstream repeats it.
• The digest of the article (724d18, p.2) says "Authorities identified the cause as a water-based explosion". The text does not say that. See accuracy check A6. This is a digest error that was not caught, because the article was never deep-read.
• Shortcuts: acceptable, with the one uncaught digest error in A6.

3. Missed material (from my own reading of the digests, deep reads and research notes)

The facts.json, claims.md, reviews, arbiter and primer missed or under-used these items:

1. Series wiring of the example generator, which could cut the cell voltage to about 1.25 V. [0049] allows the power source to be connected to the different chambers "in series or parallel" [read: 0092540 text, line 882]. Example 1 says only "applying 5 volts to a generator comprising four parallel chambers" [read: [0058]]. - If the four chambers were wired in series, each chamber's gaps would see about 5/4 = 1.25 V. That is barely above water's 1.23 V reversible decomposition voltage and below the 1.48 V thermoneutral voltage. - Every reviewer and the primer assumed 5 V per gap. That assumption is behind the "25% LHV round trip" and the "5 V is far above 1.23 V" arguments. - At about 1.25 V, ordinary electrolysis would be very slow, and making 35 psi of gas would be hard to explain. The 25% efficiency figure would also be wrong. - This does not rescue the bond-angle mechanism, but it is an open ambiguity in the patent's own example that bears on Claim 3 and on the electrolysis reading. It should be stated, not assumed away.
2. The inventors' own applications list implies that the product contains free oxygen. The 0092540 applications list includes "[0146] Firefighter Breathing Apparatus", "[0155] Tankless Underwater Breathing Apparatus", "[0191] Atmospheric Oxygen Enrichment" and "[0115] High Purity Gases" [read: 0092540 lines 1237-1406]. - The deep reader flagged that these entries "suggest the inventors expected the product to contain free O2". - None of the reviews, the arbiter or the primer uses this. It is the inventors' own text bearing on Claim 2: the product is not water molecules alone. It belongs beside the WO 2010/039283 concession. - Facts.json lists the applications only as a curiosity.
3. Pulsed-DC "intrinsic frequency". The physicist writes that "a DC (5 V) supply drives no oscillation ... no oscillating source is identified". The patent does describe an oscillating source: - pulsed DC from a MOSFET array or SCR networks, at a duty cycle of 0.005-0.5, e.g. 2 ms on and 98 ms off, which is a 10 Hz repetition rate [read: [0026]]; - AC "from a standard outlet" [read: [0024]].

So the "intrinsic frequency" has a concrete meaning: the pulse rate. The verdict is not affected, because the energy-scale argument applies to any frequency in this range. The physicist's statement needs correcting, and the primer should say what the "intrinsic frequency" most plausibly is. This pulsed-DC, close-gap design is also the link to the Meyer lineage that the Investigator noted. 4. EM-radiation embodiment. [0050]-[0051] claim that the angle can be changed by radiation of 0.1-100 cm wavelength, or 0-5 GHz (1.1 and 2.2 GHz named), applied as "pulse width modulated direct current of unequal pulse spacing". This is a second stated mechanism. - The reviewers mention it only as a units inconsistency. - Claims.md and the primer omit it. - The mathematician's note on water's dielectric absorption near 20 GHz is the relevant textbook point. 5. Raftery's dissent is missing from the primer. Dan Raftery, a Purdue chemist, is the only outside scientist on record. He said the application did not look viable and appeared to be electrolysis [web: VC Star 21 Jun 2010, via Investigator]. The arbiter keeps him; the primer drops him. - The primer should keep him, with the arbiter's caveat that he worked from news reports, not the apparatus. 6. The origin of the process and the investor's belief are not in the primer. The facts missing are: - Stehl's process was pitched from 2001 (BGX Technologies and Rainbow Technologies, Saranac Lake, NY). - Timothy A. Larson, the founder, bought the equipment in 2005 and set up Realm with his sons because he "believed in the science". That is the prosecutor's characterization, and AP repeats it. - The earlier "Rainbow Technology patents" have not been identified. - The primer's legal-record paragraph mentions Rainbow Technologies only as the fraud context. The founder's good-faith belief is pertinent and is recorded only in the mathematician's review. 7. The incident count and the 2008 detail. - Fire Engineering (2014) cites a Strategic Sciences/EPA report: "at least four explosions", including two unreported ones in 2008. - EPA POLREP #2 also mentions an apparently unreported 2008 explosion. - The Simi Valley Acorn says the December 2008 tank "imploded due to mechanical valve failure", and Larson had burns. - The VC Star says the 2010 container was being used "to heat water in an apparent attempt to break hydrogen and oxygen apart" (the authorities' characterization). - The arbiter and primer list only Dec 2008, June 2010 and Aug 2011. The "Safer: Tested by events" verdict rests on this record, so the full count should be in it. 8. A US maintenance fee was paid in March 2014. - The 4th-year fee on US 7,793,621 was paid on 2014-03-12, and the patent lapsed only in 2018 [web: Google Patents, via Investigator]. - The arbiter's absence-of-evidence analysis says "someone was still paying in 2013". The record actually shows payment in 2014, after the EP withdrawal and the parent's abandonment. - This strengthens explanation 2's "Against" point and the "payer records" lead. 9. The 2015 Saranac Lake garage find, of hazardous and reportedly radioactive items on Stehl's lot (an NCPR snippet), is in the arbiter's notes but not in the primer. It is a lead on what the original process used. 10. The Silver City, NM slag-pile application is a stated application of Stehl's energy source in the DOJ release. It is not carried into the arbiter or primer. It is minor, but it is part of the claims made for the process.

4. Accuracy spot checks (against the extracted text and page images)

5. Retention

The question here is whether pertinent material was dropped from the arbiter or primer because a reviewer judged it negatively. The material in the inventors' favour was checked first:

• The arbiter and primer keep the contrarian's "strongest honest case" in full: oxyhydrogen engines are real, the three runs are consistent, the inventors published non-zero readings, and the operation filled cylinders.
• Both keep the US grant, with its caveat.
• Both keep R2D2/H2Plus as related work.
• Both keep the tungsten observation as "the one observation mainstream expectation does not clearly cover".
• Both keep the DMCA removal as an unexplained item, neither dismissed nor inflated.
• Both keep the prosecutor's "not alleging ... fraudulent scientific process".
• Both keep the unsampled cylinders, recorded as an evidence loss and not as proof.
• The absence-of-evidence analysis (arbiter section 6) is thorough and keeps suppression and moved-abroad as live explanations.

No retention defect of the "dropped because judged negatively" kind was found.

Lesser retention gaps, where material was dropped for space rather than because of a negative judgement:

• The founder's "believed in the science" and the 2001-2005 origin of the process are missing from the primer (item 3.6).
• Raftery's outside opinion is missing from the primer (3.5). Under SOURCE HONESTY, evidence against a claim must also be stated.
• The full explosion count is missing from both the arbiter and the primer (3.7).
• The applications list is not used as the inventors' own implicit characterization (3.2).

These are fixable at MINOR level.

6. Process

• Source tags. They are present and honest throughout. [memory] confidence is stated. Investigator material is tagged "[web, via Investigator]" where the reviewer did not re-open it, which is good practice. One gap: the primer's line on Stanley Meyer and Yull Brown (section 6, "Related work") carries no tag.
• Web checks.
• Physicist: 2 used, both failed and said so.
• Mathematician: 2 used (the WO paragraphs and the barrier to linearity).
• Contrarian: about 3 (the barrier, the adiabatic flame temperature, the IMR range).
• Arbiter: 4 of 8.
• The verdict-bearing medium-confidence facts were checked (barrier to linearity; flame temperature) or shown not to be load-bearing (relaxation time, tap-water resistivity).
• Two arbiter items remain unverified because of fetch-tool limits: the text of granted claim 1, and the "Hyundai GX 390" wording in the WO text. The second is very likely a summarising-tool artefact, since the patents say "Honda GX 390". It is correctly recorded as an open question, not as fact. Both should go to the next Investigator run.
• Verdict scale. The arbiter's split verdicts are clear, but two labels are not on the Step 2 scale:
• "Tested by events, not supported" (Claim 3, "safer"). Use "Tested, not supported", with a note that the tests were uncontrolled incidents, not a designed experiment.
• "not supported as stated" (Claim 5, tungsten). The scale's nearest fit is "Never properly tested", with the calculation note that bulk melting exceeds the oxyhydrogen adiabatic limit, which the arbiter already gives.
• Otherwise the scale is applied consistently. The arbiter is right to change the physicist's "Tested, not supported" on Claim 4 to "Never properly tested", because the runs were uncontrolled.
• Step 1 not updated. Arbiter section 5 lists seven corrections to claims.md and facts.json. Neither file has been updated: facts.json still says the patent is "not stated to have been granted", and claims.md item 4 still says "near zero". Per the MINOR remediation, they should now be applied.
• Primer against STEP 4. All nine sections are present and in order. The worked example is step by step with ASSUMED values flagged, and example.py runs. Five questions are given, with answers. Reader-profile lapses:
• "Alford plea" is used undefined in section 6. The arbiter and the research note define it: a guilty plea entered without admitting guilt.
• "DMCA" (Digital Millennium Copyright Act takedown notice) is undefined.
• "WO"/PCT (international patent application) is used without explanation.
• "POLREP" (EPA pollution report) appears in Further reading undefined.
• "Adiabatic flame temperature" is used without a one-line definition: the temperature a flame reaches if no heat escapes.
• "psi gauge" versus absolute is explained only in passing.
• Lean-limit point. The contrarian noted that 1-2.5% H2 in air is below hydrogen's lean flammability limit (about 4% [memory, medium]). This strengthens the arbiter's conclusion that the O2 readings imply an oxidiser-bearing gas, negligible power, or residual gasoline. The arbiter's oxygen-budget row does not mention it. It is worth one sentence in the arbiter and in primer answer 4.

7. Remediation actions

Deep reads (most important first)

1. Ventura County Star, 4-page copy (slug Man-killed-by-explosion-in-Simi-Valley-industrial-area-724d18), pp.1-3, Sonnet. Transcribe every attributed statement with its speaker: Oatman, the authorities, "Authorities were told", the SWAT team, Debbie Shell, and the witnesses. Settle exactly what the article attributes to whom on the cause and on "extracting hydrogen from water". Note whether any company representative is quoted.
2. US 2009/0092540 (slug US20090092540-methold-and-apparatus-of-modifying-bonding-angles-of-molecules-97ca8d), pp.16-20 ([0024]-[0026], [0049], [0058]-[0059]), Sonnet. Answer three questions. How were the four chambers of the example generator connected (series or parallel), and what voltage did each gap see? Was the 5 V continuous DC, pulsed (and at what duty cycle and repetition rate), or AC? What exactly is the "intrinsic frequency" of the power source in each embodiment?
3. US 2009/0092540 (same slug), pp.20-22 (applications [0063]-[0223]), Sonnet. List every application entry that presupposes a particular composition of the product: free O2 (breathing apparatus, oxygen enrichment), free H2 (hydrogenation, "high purity gases"), redox, or a non-combustive fuel. Quote them, as the inventors' own implicit characterization of the vapor.
4. US 2009/0090312 (slug US20090090312-alternative-fuel-engine-7fd651), pp.28-32 ([0075]-[0090] and claims 1-23), Sonnet. Transcribe the CIP's claims and the engine-integration text. State whether air is admitted in every claimed configuration, and whether any engine data appear beyond Figs. 11-13. This is for comparison with the granted claim 1 of US 7,793,621, which the Investigator reports as requiring metered air and a fuel.

facts.json

• 0092540 and 0090312: add that the CIP was granted as US 7,793,621 B2 (14 Sep 2010; 4th-year fee paid 2014-03-12; lapsed 2018). Add that the parent was abandoned on 2013-09-03 and EP 2209550 was deemed withdrawn on 2013-07-05 [web, via Investigator]. Add the series/parallel ambiguity in the example and the pulsed-DC "intrinsic frequency" (after deep read 2). Promote the applications-list O2 entries from curiosity to evidence bearing on composition (after deep read 3).
• Both article entries: correct the attribution to "Authorities were told it was a water-based explosion ...; the cause was under investigation". Say that the source is unnamed. Remove "Authorities attributed".

claims.md

• Item 1: call the frequency the inventors' "RC resonating frequency", meaning an RC time-constant (relaxation) frequency with its unit unstated. Add the pulsed-DC repetition (2 ms on, 98 ms off) as the stated "intrinsic frequency". Add the EM-radiation embodiment (0.1-100 cm, 1.1/2.2 GHz) as a second stated mechanism.
• Item 4: separate the Honda claim (patent text only, no data) from the FJ-40 runs (Figs. 11-13). Replace "near zero" with the plotted vapor-phase values: CO2 0.2-0.6%, CO 0.2-0.4%, O2 20.0-20.8%, and an O2 step at engine-off. Add the oxygen-balance failure in the gasoline phase.
• Item 6: correct the attribution as above. Add that the victim was co-inventor Tyson Larson, 28 [web]. Add the later Sylmar explosion (9 Aug 2011) and the reported count of at least four explosions (Fire Engineering, citing a Strategic Sciences/EPA report).
• Header: replace "two US patent applications ... both published" with the grant and abandonment status.

arbiter.md: add a "QE revision" section

• Claim 3 ("safer") verdict label: "Tested, not supported" (uncontrolled incidents, not a designed test). Claim 5 (tungsten): "Never properly tested; bulk melting is excluded by calculation for an oxyhydrogen flame". These are label fixes, not changes of substance.
• Claim 3: add the series-wiring caveat. The 25% round-trip figure and "5 V across each gap" assume parallel wiring. If the chambers were in series, about 1.25 V per gap would make the electrolysis reading slower and the gas output harder to explain. Neither case is stated in the patent.
• Claim 2: add the inventors' applications entries (breathing apparatus, oxygen enrichment, high-purity gases) as their own implicit indication that the product contains free O2.
• Claim 4 oxygen budget: add that 1-2.5% H2 in air is below hydrogen's lean limit of about 4% [memory, medium], which strengthens "oxidiser-bearing gas, negligible power, or residual gasoline".
• Claim 6: correct the attribution ("Authorities were told"). List the full incident record: 2008 (one or two events; tank implosion due to valve failure, per the Acorn), June 2010, Aug 2011, and Fire Engineering's "at least four".
• Section 6, explanation 2: correct "still paying in 2013" to include the US maintenance fee paid on 2014-03-12.
• Physicist correction: the patent does describe an oscillating source (pulsed DC with MOSFET/SCR switching, or AC mains). The energy-scale verdict is unaffected.

primer.md

• Section 6 "Material kept": add Raftery's opinion (with its caveat), the 2001-2005 origin, and the founder's belief in the science (prosecutor's and AP's characterization). Add the applications-list O2 entries, the incident count and the Saranac Lake garage lead.
• Section 6, Claim 6 row: correct the attribution.
• Section 5, step 5: add one sentence noting the example does not say whether the chambers were in series (about 1.25 V per gap) or parallel (5 V). State which the worked example assumes.
• Section 4 table and section 7: name the "intrinsic frequency" as the pulse or mains frequency, and mention the GHz radiation embodiment in one line.
• Define Alford plea, DMCA, WO/PCT, POLREP and adiabatic flame temperature. Tag the Meyer and Brown line [memory, confidence].
• Answer 4: add the lean-limit sentence.

Investigator (next run, already partly listed in sites.md)

• Open WO 2010/039283 [0074]-[0078] directly to settle "Honda" versus "Hyundai" GX 390, and US 7,793,621 claim 1 verbatim.
• Open VC Star, 18 June 2010 ("Authorities identify man killed ..."), which may name who told the authorities it was "a water-based explosion".

---

Recheck

Date: 2026-10-04. Re-read facts.json, claims.md, arbiter.md (incl. new section 10 "QE revision"), primer.md against the Section 7 remediation list. Method: targeted greps and reads of the current files.

Remaining defects: none in the documents. The three open Investigator items are recorded as unknowns, not asserted as facts, and are not defects.

Final: PASS