Working Hypothesis · Open for Revision
One candidate embodiment of a working Hum — built around back-EMF recovery, the most independently-convergent and cheapest-to-falsify of the five mechanism clusters we read out of 768 patents and 35 experiments. This is a living document. Every section links to the experiments that would confirm or refute it.
An earlier version of this page claimed all 768 patents converge on one recipe — non-linear + resonance + pulsed. Reading the actual claims, not just keyword counts, corrected that: only about 2% of patents combine all three. The real signal is narrower and stronger — five mechanism templates that unrelated inventors keep arriving at independently, decades and continents apart, with no shared citations:
Capture the inductive-collapse spike and feed it back. Bedini and a dozen unrelated inventors arrived at it independently — and it is the single cheapest claim to falsify (a battery-to-battery coulomb count). This prototype is built around it.
Asymmetric vacuum-fluctuation cavities feeding a rectifying junction. Reached independently by three or more unrelated inventors — the most internally consistent anomalous family.
Tune an electrical or acoustic resonance to split water below the Faraday minimum. US, Russia, Chile, Japan — 30 years, no shared citations.
Hydrogen loaded into a nickel or palladium lattice, then "rung" by current, magnetic, acoustic, or THz stimulation for claimed excess heat.
A rotating (E×B) field driving low-energy fusion collisions — the same mechanism filed independently under more than one patent class.
The prototype below is one candidate embodiment — built primarily around back-EMF recovery, the cluster that's both independently convergent and the cheapest to settle with a single meter. The pulsed drive, bifilar core, and resonance tuning are design choices in service of that loop — not universal laws of the corpus.
Pulse Generator
555 Timer + IRF540N MOSFET
40 kHz, 10–15% duty cycle, ~100ns rise time
The Spark drives the recovery loop: a fast, low-duty-cycle pulse so every switch-off produces a sharp back-EMF spike for the Bridge to capture. The sharp rising edge (~100ns) also spreads energy across harmonics, in case the Core's non-linearity contributes. A 555 timer is cheap, tunable, and well-understood; the IRF540N handles the current without significant switching losses at 20–100 kHz. Duty cycle in the 5–20% range keeps the on-time short so the collapse spike is large.
Medium — duty cycle and frequency are the main unknowns. The patent literature suggests 10–15% but this needs per-system tuning.
Non-Linear Element
Bifilar-wound coil on FT-50-43 ferrite toroid
50–100 turns bifilar (series-aiding), driven near saturation
A bifilar-wound ferrite toroid (Tesla's 1894 configuration, US512340) driven toward saturation. The bifilar winding adds large inter-winding capacitance, which shifts the self-resonant frequency and gives the Tuner a built-in LC resonance to lock onto; pushing the core toward saturation adds a sharp non-linearity that generates harmonics from the pulse. Non-linearity is a design choice here, not a law of the corpus — reading the claims, most patents show no non-linear element at all — but it is cheap to include and easy to measure. Mix 43 ferrite keeps the core's own resonances (20–250 MHz) well above the operating band, avoiding confounding effects.
High — bifilar geometry definitively shifts SRF vs conventional winding. The question is whether this shift matters for energy conversion.
Resonance Lock
Variable capacitor + feedback tap
Tuned to the Core's self-resonant frequency (typically 1–10 MHz for a 50-turn bifilar on FT-50-43)
Resonance is a tuning lever, not the corpus's dominant pattern — an earlier read called it "the most common element," but reading the actual claims, most patents show no resonance at all. So we treat it as an optional amplifier, not the effect itself. The Tuner sweeps a variable capacitor against the Core's parasitic capacitance to find and lock maximum impedance, and a feedback tap lets the pulse frequency track that resonance as temperature, load, and component aging drift.
Medium — we know resonance matters, but the optimal tuning strategy (fixed vs adaptive) is unresolved.
Energy Recovery
Fast-recovery diode bridge + storage capacitor
UF4007 diodes (1A, 75ns recovery), 100µF electrolytic storage cap
This is the heart of the prototype. Back-EMF recovery is one of the five convergence clusters — the "capture the collapse spike and feed it back" template that Bedini and a dozen unrelated inventors arrived at independently across decades, and the single cheapest anomalous claim to falsify. On each switch-off the Core kicks back a spike that can exceed the drive voltage 10–50×; a conventional flyback diode burns it as heat, while the Bridge routes it into a storage capacitor. Fast-recovery diodes (75ns) are essential to catch the nanosecond-scale edge. Whether net recovery ever exceeds input is exactly the question a battery-to-battery coulomb count settles — which is why this cluster tops the buildable list.
Low-Medium — recovery circuit topology is the area with the most variation across patents. Multiple approaches may work.
Output
USB 5V regulator + LED indicator
5V / 500mA output (2.5W), green LED = producing, red LED = consuming
The Load is how you know if it's working. A simple USB regulator converts whatever the Bridge captures into a usable 5V output. The LED indicator provides instant visual feedback: green means the system is outputting net energy to the load; red means it's consuming more than it produces (which is the expected state until the system is properly tuned). This isn't the final form — it's the measurement stage. If the green LED stays on with no external power input, that's the result everyone is looking for.
N/A — the load circuit itself is straightforward. The question is whether the upstream system produces enough energy to light it.
| Component | Part | Est. Cost |
|---|---|---|
| Spark | 555 timer + IRF540N MOSFET + passives | $5 |
| Core | FT-50-43 toroid + 30 AWG magnet wire | $10 |
| Tuner | Variable capacitor (air-dielectric) | $8 |
| Bridge | UF4007 diodes (x4) + 100µF capacitor | $3 |
| Load | USB 5V buck regulator + LEDs | $4 |
| Measurement | NanoVNA-H4 (if you don't have one) | $30 |
| Misc | Perfboard, wire, solder, connectors | $10 |
| Total (without NanoVNA) | ~$40 | |
| Total (with NanoVNA) | ~$70 | |
This is a falsifiable hypothesis. Here's what would make us revise each section:
Every experiment you run either strengthens or weakens a section of this diagram. Both outcomes move the project forward. The only thing that doesn't help is not building.
Three concept renders
AI-generated from the component specs above. Hover any image for the prompt used to generate it — fork it, improve it, generate your own.