When Did the Heat Happen?

1. The Heat Is Something You Calculate, Not Something You Read Off a Gauge

Imagine you’re asked to measure how much extra heat a circuit board is producing beyond what its power supply puts in. You wouldn’t get that number from a single sensor. You’d log temperature, current, and voltage over time, subtract out a baseline, correct for how heat leaks out of the box, and integrate the result. Every step — sampling, timestamping, subtracting — is a place where error can creep in before a human ever sees an answer.

That’s exactly the situation in solid-state fusion (SSF), the field once known as “cold fusion.” For thirty years, researchers have argued over a quantity called excess heat, the extra power a small experimental cell produces beyond the electrical power fed into it. If a cell is putting out more energy than went in, and nothing chemical explains the surplus, that surplus is the signature a nuclear reaction would leave behind. But excess heat is never a raw reading. It’s the output of a calorimetric model, essentially a small data-processing pipeline, built from temperature, current, and voltage streams. Every disputed watt has already passed through sampling, timestamping, and arithmetic. That means the argument over whether the effect is real is, in large part, an argument over instrumentation, which is to say, an argument in your field, not in nuclear physics.

2. The Experiment That Would Actually Settle It Is a Timing Problem

Here’s the sharper version of the question. The single most convincing thing SSF researchers could show is a coincidence: proof that when the heat goes up, a matching burst of nuclear byproducts — neutrons, gamma rays, or helium — shows up at the very same moment. Real fusion has to produce both energy and characteristic particles together. If you can show the two rising in lockstep, “coincidence to a known uncertainty,” you have a real result. If you can’t, a suspicious pile of heat with no matching particles is a red flag, not a discovery.

Showing that two events happened “at the same time,” to a stated precision, is a classic embedded-systems problem: it’s the same kind of challenge as making sure two microcontrollers on a shared bus agree on when a signal fired. You need two independent instruments (the heat-measuring circuit and a particle detector) sharing one trustworthy clock, so a claim like “the particles appeared as the power rose” comes with an actual error bar instead of just a story. A better calorimeter alone can’t get you there. What’s needed is triggered, synchronized data acquisition, the kind of clock discipline this field has almost never built to a real engineering standard.

It’s worth being honest about what this can and can’t prove. A cleaner instrument can’t make a heat signal real if it isn’t, and a real heat signal isn’t automatically a nuclear one. What good timing and capture can do is guarantee a trustworthy answer either way: a well-characterized anomaly worth investigating, or a clean null result with a stated sensitivity attached to it.

3. Why “Pretty Good” Instruments Still Weren’t Good Enough

Consider the most useful failed experiment in the field’s history, useful because the team was unusually honest about exactly where their instrument fell short. In 2016, a small group at ReResearch LLC tried to replicate a well-known dual-laser excess-heat claim and found nothing real: 6.1 ± 21.6 milliwatts of excess power averaged across 231 trials, on a cell drawing about 10 watts of input power.¹

Read those numbers the way an instrumentation engineer would. The effect they were hunting for, an original claim of about 100 to 300 milliwatts, was a percent or two of the input power. What the experiment could actually resolve was a residual near one part in a thousand of the input, with an uncertainty nearly four times larger than that number. The hardware noise floor, as the next figures show, was itself as large as the effect they were trying to see. And that noise wasn’t a math mistake, it was hardware: artifacts of 130 to 460 milliwatts showed up just from calibration and setup, and mechanical shifting of the temperature probes added another 135 to 324 milliwatts, enough to throw off the sensed temperature by up to about a degree Celsius.¹

The lesson generalizes past this one experiment: a clean analysis pipeline can’t fix a noisy front end. If your sensor mount can wiggle, or your reference temperature drifts, you’re measuring your own hardware breathing, not the thing you set out to test. Milliwatt-scale signals riding on a baseline that wanders by hundreds of milliwatts are answered with better fixturing, guarded references, and calibration that’s traceable to a known standard, the same toolkit you’d reach for chasing any small signal against a noisy background, in any field.

The payoff of fixing the hardware is concrete: shrink that noise floor by a factor of ten, and the 2016 null result stops being ambiguous. What was buried becomes something you can actually confirm or rule out. A negative result only means something once your instrument is sensitive enough to have caught the effect if it were there.

4. Keeping the Experiment From Wandering While You Measure It

There’s a second reason this problem needs engineers, not just better sensors. The cells being tested aren’t quiet, stable samples sitting on a shelf. A working electrode is being actively driven by current and voltage, and its material properties, in particular how much hydrogen has soaked into the metal, drift and shift unpredictably while the experiment runs. If that internal “loading” isn’t held steady, the experiment isn’t reproducible no matter how good your later analysis is.

This is a feedback-control problem, the same category as a cruise control system holding a car’s speed steady, or a thermostat holding a room’s temperature. You measure a quantity in real time, compare it to the target, and continuously adjust. In SSF, one standard signal engineers can track this way is the electrode’s electrical resistance, which changes predictably as it soaks up more hydrogen. Doing that control loop with a general-purpose computer running a normal operating system isn’t good enough, because you need guaranteed, repeatable timing: a response that happens within a fixed window, every single time, not “usually pretty fast.” That’s what a real-time operating system (RTOS) is built to guarantee. Pair that steady, disciplined control with careful record-keeping, saving every raw measurement and every step used to process it, and any strange result becomes something you can trace back and explain, rather than something you just remember happening.

5. What Engineering Gets Back From Taking This On

None of this is a favor to SSF researchers. It’s a genuinely hard, realistic test bed, and the return is real. Detecting a microwatt- to milliwatt-scale signal riding on a multi-watt input, holding a reading steady over weeks rather than seconds, and synchronizing very different kinds of sensors, thermal, electrical, and nuclear, that operate on completely different time scales, is a legitimately difficult combination. Methods good enough to solve it here, precise clock synchronization across separate instruments (the kind of protocol, IEEE 1588, that already keeps networked devices agreed to within a millionth of a second), and long-term data provenance, transfer directly to any engineering problem chasing a small signal against a noisy, shifting background.

There’s also a cultural lesson worth borrowing. Safety-critical software, the code that flies airplanes or runs industrial control systems, follows standards like DO-178C and IEC 61508, which scale how much rigor and independent testing a system needs to how bad it would be if that system failed. Applying that same instinct, effort matched to the stakes, to scientific instruments in general, where a bad measurement means a wrong published number rather than a plane crash, is a real and underused idea. So is bringing in an independent team to check the instrument chain without ever weighing in on what the physics results mean; it’s one of the strongest safeguards against a team unintentionally finding what it hoped to find.

This isn’t just an academic argument, either. In 2023, the U.S. Advanced Research Projects Agency–Energy (ARPA-E) committed roughly $10 million across eight research teams, including groups at MIT, Stanford, and Lawrence Berkeley, specifically to bring modern instrumentation to bear on these claims.² A few years earlier, a Google-funded effort published a careful negative result in Nature, and, more usefully, pointed at exactly where the older experiments had been too crude to draw any firm conclusion: the materials science of how much hydrogen a metal can hold, and the measurements built around it.³ That’s an open invitation, from people with no stake in the outcome, to bring better engineering to a question that’s stayed unresolved mostly for lack of it.

The Bottom Line

“When did the heat happen?” is, underneath the physics dispute, an ordinary measurement question, the kind engineers make trustworthy every day: get two instruments to agree, precisely, on when something occurred. The honest deliverable isn’t a verdict on whether cold fusion is real. It’s a timestamp with an error bar attached, a coincidence window that either brackets a genuine correlation or rules one out at a stated level of confidence. That measurement has essentially never been posed to this field with instruments actually built to the standard the question deserves. It’s answerable. It just hasn’t, mostly, been asked properly yet.

Editorial note: This primer presents an introductory synthesis of solid-state fusion's relationship to embedded systems, RTOS, and FPGA engineering, leveled for first-year engineering students. The underlying nuclear claims of SSF/LENR remain scientifically contested. Readers are directed to primary experimental literature, and to the companion expert-level primer, for a more detailed and fully sourced treatment.

References & Footnotes

1. The 2016 replication attempt and its itemized calibration and probe-displacement artifacts are drawn from a published null result testing a dual-laser excess-heat claim. Established as a reported measurement; the underlying phenomenon it failed to detect remains contested. ↩
2. U.S. Advanced Research Projects Agency–Energy (ARPA-E), 2023 program funding low-energy nuclear reaction research across eight university and national-lab teams. Established (a matter of public record; does not itself validate any underlying claim). ↩
3. C. P. Berlinguette et al., “Revisiting the Cold Case of Cold Fusion,” Nature 570 (2019): 45–51. Did not reproduce excess heat, but identified extreme-loading materials science and its instrumentation as the field’s weakest, most under-studied link. Established. ↩
4. The heat-helium correlation reported by Miles, McKubre, and co-workers (energy-per-helium ratios landing within roughly an order of magnitude of the 23.8 MeV expected from deuteron fusion to helium-4) is presented in the source article as reported, not established; it is contested on calorimetry and atmospheric-helium-contamination grounds and is omitted from the numeric detail here in favor of the general “coincidence measurement” framing in Section 2. ↩

Introduction: A Measurement Chain, Not Just Nuclear Physics

Start with the instinct every real-time engineer carries into a lab: a measurement is worth exactly as much as the chain that produced it and the clock that stamped it. Solid-state fusion (SSF), the field once called cold fusion and now more precisely condensed matter nuclear science, has spent thirty years arguing about a number that lives entirely inside that chain. The excess heat at the center of the controversy is a quantity nobody reads off an instrument; it is computed from raw temperature, current, and voltage records through a calorimetric model: baseline subtraction, a calibration curve, a heat-loss model, and an integral over time.¹ Every disputed watt passes through sampling, timestamping, and arithmetic before anyone sees it.

That should get an embedded engineer’s attention, because it means the disputed object is one your field already owns: the measurement chain, not the nuclear physics. And the sharpest version of the SSF question turns out to be a timing problem in disguise. The most decisive experiment anyone can run is a coincidence measurement, showing that an energy-matched flux of nuclear products rose when, and in proportion to, the heat. A better calorimeter alone will not settle it. Getting two independent instruments to agree on when something happened, to a stated uncertainty, is the daily work of triggered data acquisition, and this field has almost never had it built to standard.

The honest frame is narrow and worth stating up front. Better instrumentation cannot make a heat signal real, and a real heat signal is not by itself a nuclear one. What deterministic capture and disciplined timing can do is return a trustworthy answer either way: a well-characterized anomaly, or a clean null with its sensitivity attached. The most useful thing embedded systems can contribute here is not endorsement of a mechanism. It is a chain that cannot quietly lie.

Section I — The Number Is an Integral

Consider the most instructive negative result in the literature, instructive because its authors itemized their hardware error budget as honestly as any paper in the field. In 2016 a small group at ReResearch LLC attempted to replicate a well-known dual-laser excess-heat claim and found nothing: 6.1 ± 21.6 mW of excess power averaged over 231 trials, against a cell input near 10 W.² Excess heat here is simply the cell’s measured power output minus its electrical power input, the surplus a nuclear source would have to supply. Read those numbers as an instrumentation engineer. The effect being hunted, the original claim of roughly 100 to 300 mW, is a percent or two of the input; what the experiment could actually resolve was a residual near the 10⁻³ level, and its uncertainty was nearly four times any central value. And the confounds the authors itemized were not analysis mistakes. They were hardware: artifacts of 130 to 460 mW seen during calibration and setup, and 135 to 324 mW from mechanical displacement of the temperature probes, enough to shift the sensed temperature by 0.5 to 1.2 °C. The floor that swallowed the signal was made of physical apparatus.

This is the boundary that matters most for anyone arriving from software. A validated analysis pipeline, however rigorous, cannot lower a hardware error floor; that limit is set by the front end, the sensor mounting, the reference, and the thermal design. Which is precisely why the problem belongs to embedded and instrumentation engineering rather than to code review. A milliwatt-scale signal riding on a baseline that wanders by hundreds of milliwatts is a familiar adversary, answered with synchronous and ratiometric measurement, rigidly fixtured and thermally anchored probes, guarded references, and calibration traceable to a standard. The general lesson outlives this one experiment: every temperature reading rests on a reference that is itself a measurement. A thermistor drifts, self-heats, and moves if its mounting lets it; a thermocouple reads against a cold junction whose own temperature you have to know. Treat either reference casually and the drift you see is your own instrument breathing. None of this is exotic. It is the core competence of the field, aimed at a target that happens to be contested.

The payoff of aiming it here is specific. Lower the hardware floor by an order of magnitude and the 2016 null stops being ambiguous. A signal that was buried becomes a signal you can confirm or exclude. Note what that buys and what it does not: a trustworthy measurement, not a positive one. A null retires a hypothesis only when the instrument’s sensitivity clears the effect size being claimed. Below that, a null is silence, not absence.³ The whole value of a better front end is that it moves the sensitivity to where the answer means something.

Section II — The Decisive Measurement Is a Coincidence

The observation that has always carried the most weight in SSF is a correlation: in a series of palladium heavy-water cells, cathodes that produced more excess power reportedly produced more helium-4, with the energy-per-helium ratio landing within roughly an order of magnitude of the 23.8 MeV expected if deuterons were fusing to helium-4.⁴ The result has been reported by a small number of groups, principally Miles at China Lake and, independently, McKubre and co-workers at SRI, and it is contested on calorimetry and contamination grounds. It is presented here as reported rather than established. Set aside for a moment whether it holds. Ask instead what it would take to test it to a standard no reasonable critic could wave away.

The answer is a coincidence measurement, and coincidence is a timing discipline. You need the calorimetry channel and an independent product detector, high-purity germanium for gammas, time-of-flight for neutrons, mass spectrometry for helium, sharing a common and trustworthy time base, so that “the product appeared as the power rose” becomes a statement with an error bar rather than a narrative. The skeptical objection that has never been answered cleanly is about products: at that rate, ordinary deuteron-deuteron fusion would split almost evenly between its two branches and flood the cell with neutrons and tritium, while the direct fusion-to-helium-4 route implied by the heat-helium reports would instead emit a matching flux of 23.8 MeV gamma rays, and no such commensurate flux, of either kind, has reliably appeared.⁵ That objection is, at bottom, a demand for correlated bookkeeping: same events, same clock, same uncertainty budget.

Here the home field has tools the SSF literature has barely touched. Networked measurement and control has a mature synchronization standard, IEEE 1588, the Precision Time Protocol, which disciplines clocks across separate instruments to sub-microsecond agreement over ordinary Ethernet.⁶ Where that is not tight enough, the White Rabbit extension developed for accelerator timing distributes a common clock to sub-nanosecond accuracy across kilometers of fiber, and it was built precisely so that detectors in different rooms could agree on when.⁷ Time-of-flight neutron spectroscopy and pulse-shape discrimination between neutrons and gammas are real-time classification tasks that FPGAs already perform in mainstream nuclear instrumentation, with deterministic latency and honest deadtime accounting. The instruments a decisive SSF coincidence experiment would need are, individually, off-the-shelf in the DAQ world. What has been missing is the integration: one disciplined time base tying heat and product into a single triggered record.

Picture the record such an experiment should produce. A product detector supplies the trigger: a high-purity germanium gamma event, or a neutron flagged by pulse-shape discrimination, timestamped in the FPGA at the moment of detection. That timestamp is compared against the calorimetric power on the same disciplined clock, so “the product appeared as the power rose” has a defined reference time rather than a narrator. Around each trigger you open a coincidence window whose width is set by the physics you expect and the timing budget you can defend, and you estimate the accidental rate from the out-of-window sidebands, so any claimed correlation is quoted against the background of chance coincidences. Detector deadtime is logged per channel and corrected, because a busy detector that failed to record is indistinguishable, after the fact, from a real absence of counts. What comes out is a table of timestamped events carrying an accidental-coincidence rate and a stated per-channel sensitivity, which either shows a correlation above chance or sets a numeric limit on one, rather than a story about heat and helium rising together.

The category discipline has to hold even here. A clean coincidence, product flux rising in lockstep with power, would be a genuine anomaly and a serious problem for the standard picture. It would still not, by itself, prove a nuclear origin, and the article makes no such leap. It would move the question from “is the number trustworthy” to “what mechanism could do this,” which is exactly where a live scientific problem should sit.

Section III — Determinism as Experimental Control

There is a second reason this field needs embedded engineering, and it runs deeper than acquisition. A working electrolytic cathode is a driven, far-from-equilibrium system, anything but a quiet sample, with steep loading gradients and a surface that fractures unpredictably under transient stress. Its “conditions,” the loading fraction, current density, and temperature that any claimed effect is supposed to depend on, are only as well-defined as the real-time loop holding them. If the stimulus wanders, the experiment is not reproducible no matter how clean the later analysis.

Hard real-time control is the answer to that, and it is why an RTOS belongs in the apparatus rather than a best-effort script on a general-purpose operating system. The standard non-destructive loading diagnostic, the resistance ratio of the cathode, which for deuterium in palladium tracks loading (the deuterium-to-palladium atomic ratio, D/Pd) up toward the 0.85 to 0.90 range associated with excess-heat reports, is a signal you can close a control loop around in real time rather than read off afterward.⁸ Deterministic control keeps the stimulus where the protocol says it should be, and deterministic capture records exactly what the stimulus did. Pair that with provenance discipline borrowed from regulated industries, where every raw stream and every transformation is preserved so a result can be re-derived from first records, and each excursion becomes attributable and reconstructable rather than a story told from memory.⁹ This is the reproducibility lever embedded systems actually own. Not the trustworthiness of the analysis, which is a software concern, but the physical repeatability of the stimulus and the fidelity of the record.

Section IV — What the Home Field Gets Back

None of this is charity. SSF hands embedded and instrumentation engineering an adversarial testbed at the edge of the discipline’s practice, and the return cargo is real.

The measurement itself is a worst case worth having. Micro- to milliwatt signals riding on multi-watt inputs; drift that has to be held over weeks, not seconds; multi-domain triggering across thermal, electrical, and nuclear detectors that run on different physics and different timescales; and closed-loop control of a material that changes state unpredictably underneath the controller. Solve calorimetry-grade synchronous measurement and long-baseline provenance here and the methods travel directly to any field chasing a small signal against a restless background.

The traffic in standards runs both ways too. Safety-critical software culture already scales rigor to the cost of being wrong: DO-178C assigns design assurance levels A through E by the consequence of failure and demands that requirements trace bidirectionally to code and tests, and IEC 61508 does the equivalent with safety integrity levels for industrial systems.¹⁰¹¹ Those frameworks were built for systems where a fault kills someone. Porting their central idea, assurance proportional to stakes, to scientific instrumentation, where the failure mode is a wrong derived number rather than a crash, is a genuine research direction for the field that wrote them. So is independent verification and validation: the practice, formalized in IEEE 1012 and institutionalized in programs like NASA’s IV&V facility, of having a separate team confirm a system does what is claimed.¹² An instrumentation group that validates the measurement chain while never touching the physics interpretation is the strongest available guard against a self-fulfilling result, and SSF is an almost ideal case study for the method.

The institutional pull is already there. In 2023 the U.S. Advanced Research Projects Agency–Energy committed roughly $10 million across eight teams, including groups at MIT, Stanford, and Lawrence Berkeley, to test low-energy nuclear reaction claims with modern instrumentation.¹³ A few years earlier a Google-convened effort published a careful negative result in Nature and, more usefully, named where the older work had been too crude to conclude anything: the materials science of extreme loading, and the measurements around it.¹⁴ That is an explicit invitation to bring better instrumentation to an under-mapped problem, issued by people with no stake in the answer coming out positive.

Section V — Build the Chain That Cannot Lie

The most productive contribution embedded systems can make to solid-state fusion is an apparatus that returns an answer worth believing whichever way it falls, rather than a vote on whether the effect is real: deterministic real-time control holding the stimulus steady, synchronized multi-channel capture tying heat to product in a single triggered record, provenance complete enough to re-derive every number, and an independent team validating the chain end to end. Report every outcome with its sensitivity, and two of the three possible results are valuable. A coincidence that survives that scrutiny is a real anomaly and the start of a hard, interesting mechanism problem. A clean null at a stated sensitivity that clears the claimed effect finally retires a corner of the map that has stayed frustratingly grey. The only worthless outcome is the one the field has produced too often already: another number nobody outside the room can check.

“When did the heat happen?” is a measurement, of the ordinary kind embedded engineers make trustworthy every day. The deliverable is a timestamp with an error bar, not a verdict: a coincidence window that either brackets a real correlation or bounds it. That measurement has simply never been posed to this field with instruments built to the standard the claim deserves. It is answerable, and it has mostly gone unasked.

Editorial note: This primer presents a scholarly synthesis of solid-state fusion's relationship to embedded systems, RTOS, and FPGA engineering, written for an expert audience. The underlying nuclear claims of SSF/LENR remain scientifically contested. Readers are directed to primary experimental literature for empirical evaluation.

References & Footnotes

1. Excess heat in SSF calorimetry is a computed quantity, derived from raw temperature, current, and voltage streams through a calorimetric model (baseline subtraction, calibration, loss model, integration), not a directly measured one. Edmund Storms, The Science of Low Energy Nuclear Reaction (Singapore: World Scientific, 2007); standard calorimetry practice. ↩
2. M. J. Guffey, Y. Tang, and P. J. King (ReResearch LLC, Hawthorne, CA), “Attempted Replication of Excess Heat in the Letts Dual-Laser Experiment,” Journal of Condensed Matter Nuclear Science 20 (2016): 1–28. Reported excess power 6.1 ± 21.6 mW over 231 trials at ~10 W input; itemized hardware artifacts: 130–460 mW during calibration and setup, and 135–324 mW from mechanical displacement of the (Measurement Specialties 44004RC) thermistor probes, which shifted the sensed temperature by 0.5–1.2 °C. ↩
3. A null result retires a hypothesis only if the measurement’s sensitivity clears the claimed effect size; at or below the error floor, a null is absence of evidence at insufficient power, not evidence of absence. Standard statistical-power and detection-limit reasoning. ↩
4. M. H. Miles, B. F. Bush, et al., “Correlation of Excess Power and Helium Production during D₂O and H₂O Electrolysis Using Palladium Cathodes,” Journal of Electroanalytical Chemistry 346 (1993): 99–117; corroborating heat–helium measurements were reported independently by M. C. H. McKubre and co-workers at SRI International. Energy-per-helium reported as roughly consistent (within about an order of magnitude) with 23.8 MeV per ⁴He. Reported by a small number of groups and contested on calorimetry and atmospheric-helium contamination grounds; presented as reported. ↩
5. At that rate, ordinary deuteron–deuteron fusion would route roughly half its events down each of two branches, t + p and ³He + n, and flood the cell with neutrons and tritium, while the direct ⁴He + γ branch is strongly suppressed (branching ~10⁻⁶–10⁻⁷) and would emit a 23.8 MeV gamma; no commensurate flux of either kind has reliably appeared (branching per ENDF/B-VIII.0). U.S. Department of Energy, Report of the Review of Low Energy Nuclear Reactions (Washington, DC, December 2004); Storms, The Science of Low Energy Nuclear Reaction. General absence is well supported; specific positive product reports remain single-community and contested. ↩
6. IEEE Std 1588-2019, IEEE Standard for a Precision Clock Synchronization Protocol for Networked Measurement and Control Systems (New York: IEEE, 2020). PTP disciplines clocks across networked instruments to sub-microsecond agreement over standard Ethernet. ↩
7. J. Serrano et al., “The White Rabbit Project,” Proceedings of ICALEPCS 2009; and M. Lipiński et al., “White Rabbit: a PTP Application for Robust Sub-nanosecond Synchronization,” 2011 IEEE International Symposium on Precision Clock Synchronization (ISPCS). A PTP/SyncE extension distributing a common clock to sub-nanosecond accuracy over fiber, developed for accelerator timing at CERN and GSI. ↩
8. M. C. H. McKubre et al., “Using Resistivity to Measure H/Pd and D/Pd Loading: Method and Significance,” in Condensed Matter Nuclear Science (ICCF-12 Proceedings) (Singapore: World Scientific, 2006). The resistance ratio is the standard non-destructive loading diagnostic; excess-heat reports cluster at high loading, D/Pd ≈ 0.85–0.90, with high loading necessary but not sufficient. Threshold values are approximate. ↩
9. Data-integrity practice in regulated industries (ALCOA/ALCOA+: attributable, legible, contemporaneous, original, accurate, plus complete, consistent, enduring, available), reframed for instrumentation as preserving the raw stream and every transformation so a result can be re-derived. Medicines and Healthcare products Regulatory Agency (MHRA), ‘GXP’ Data Integrity Guidance and Definitions, Rev. 1 (March 2018). ↩
10. RTCA DO-178C, Software Considerations in Airborne Systems and Equipment Certification (Washington, DC: RTCA, 2011). Design assurance levels A–E scale required rigor to the consequence of failure; requirements trace bidirectionally to code and tests. Borrowed here as a principle, not a claim that SSF instrumentation currently meets it. ↩
11. IEC 61508, Functional Safety of Electrical/Electronic/Programmable Electronic Safety-Related Systems (Geneva: IEC). Defines safety integrity levels (SIL 1–4) scaling required rigor to risk. ↩
12. IEEE Std 1012-2016, IEEE Standard for System, Software, and Hardware Verification and Validation (New York: IEEE, 2017); NASA Independent Verification and Validation (IV&V) Program. IV&V is V&V performed by a technically, managerially, and financially separate organization. ↩
13. Advanced Research Projects Agency–Energy (ARPA-E), “U.S. Department of Energy Announces $10 Million in Funding to Projects Studying Low-Energy Nuclear Reactions,” February 2023. Eight teams, including groups at MIT, Stanford, and Lawrence Berkeley National Laboratory. ↩
14. C. P. Berlinguette et al., “Revisiting the Cold Case of Cold Fusion,” Nature 570 (2019): 45–51. The Google-convened team did not reproduce excess heat but identified extreme-loading materials science, and the measurements around it, as underexplored. ↩