The simplest fusion machine imaginable — rebuilt from its patent.
We have never seen this machine. Everything in the film below — the geometry, the firing sequence, the instability that killed the Z-pinch for seventy years and the flow that tames it — was rebuilt from one public document: US patent 12,219,686 B2.
Master delivered as promised: when this page first went live it showed the direction preview, with the 4K master still on the render farm — 4,380 frames, path-traced, 128 samples per pixel. What you watch now is that master: same scenes, same cameras, same frames. That is what deterministic means.
This is an independent technical reconstruction from public documents, produced by Verlet as a demonstration of method. Verlet is not affiliated with, commissioned by, or endorsed by Zap Energy. No proprietary information was used — none was available. That is the point.
Public sources only
The patent, the papers it cites, and nothing else. No speculation about unpublished performance, no claims the documents don't make. Where the patent gives a range instead of a number — the intermediate electrode's taper, 0–15° — the film cites the range and we declare our choice.
Every figure is accountable
The patent states no absolute radii, so we measured its cross-section drawing at 300 dpi and anchored the scale to the one dimension the text does give: gas valves at −50, −75 and −100 cm. The measurement closes on itself — the inner electrode comes out at 106 cm against the text's ~1 m. Every on-screen number is either measured from our model or cited to its source.
The physics is simulated, not decorative
The m=0 "sausage" instability in the film grows at the patent's published rate on a Bennett equilibrium, and the sheared flow suppresses it on the patent's own curves. An automated gate checks the model against the physics before any frame ships — the report is below, unedited.
Animation studios show you a reel. We show you the checks our model had to pass before we let it render — each one an independent computation compared against the patent's stated values and figures.
| Ampère's law | max error 0.000% over 48 independent loops (threshold 1%) | PASS |
| Bennett peak | Bθ peaks at r = 0.909 mm against a = 0.91 mm | PASS |
| Bennett profile | correlation with x/(1+x²): 1.000000 | PASS |
| Growth rate | fitted γτA = 1.0000 · patent range 0.7–1.5 | PASS |
| Wavenumber | ka = 10.0 · patent: 10 | PASS |
| Lifetime, no flow | column lost at 6.0 τA — where FIG. 5 loses it | PASS |
| Lifetime, sheared | alive past 30 τA at vsf/vA = 0.75 — where FIG. 5 keeps it | PASS |
| Shear monotonicity | γeff falls with flow: 1.00 → 0.50 → 0.30 → 0.10 | PASS |
| Seed consistency | γeff(0) = 1.000 equals the fitted rate — one seed ties FIG. 5 to the growth fit | PASS |
| Pinch current | I = 300 kA, as stated | PASS |
| Pinch radius | a = 0.91 mm, as stated (FIG. 23) | PASS |
τA = 2.61 ns — the film runs the collapse at 1:200,000,000 so you can watch it. The gate exits non-zero on any failure: the piece does not ship unless every check passes.
This film was built at arm's length, from a patent, on purpose. Radii read from a perspective drawing carry ±10%. The firing sequence follows the patent's schematic panels, not a measured discharge. That is the honest ceiling of public documents — and a commissioned piece starts where this one had to stop.
Your geometry, exact
Drawing-measured ±10% becomes your engineering geometry, to the part. Same cameras, same film — every dimension now yours.
Your discharge, not a schematic
Current traces, gas timing and diagnostic profiles replace the patent's published figures. The sequence on screen becomes your machine firing.
Your engineers, in the loop
Unlimited revisions on fast previews, with the people who built the machine marking what's wrong. What they approve is exactly what renders.
Your name on it, in any language
Annotations are composited over the frames, never burned in. A second language, your branding, a 90-second cut — variants at marginal cost, without touching the render.
This piece used the patent, and nothing else. Imagine it with the data only you have.
If your machine deserves this treatment, this is what we do — for the machines we're allowed to see.