A spiral twists what escapes it
The Kozyrev mirror, rebuilt as an instrument that is hard to fool. A real wave equation runs in your browser, every number on this page is measured from the field, and the panel on the right checks the instrument while you use it.
The verdict
Real effect, different reason
A spiral does twist the waves that escape it, by ordinary reflection. In equilibrium the twist is zero, and the old +19.05 was an integrator artifact. The mechanism is textbook prior art.
Waves that leave the spiral leave turning. Seal the box (closed, lossless, a real source) and the net twist is exactly zero. A hot spiral radiating into a cold outside also twists what it emits, like a tiny heat engine.
At one temperature inside and out, the net angular-momentum flux is at most 1.3e-4 of the gross. In this code that check is close to an identity, so the equilibrium result also rests on the physics literature.
The lab's earlier "robust" number existed only with a time step past its stability limit. At a stable step it falls from 18.94 to about 1e-4. You can reproduce the kind of bug with Unstable step below.
Spiral reflectors that twist waves are established in radio, optics and acoustics. Thermal torque on chiral bodies needs a temperature difference (Reid et al. 2017; Milton et al. 2024–25). No novelty is claimed.
This episode tested the lab's own earlier claim that a spiral twists fluctuations into net angular momentum. The Why Files episode “Bending Time: The Successful Time Travel Experiments using Kozyrev Mirrors” (June 2023, the episode) presents a different mechanism: the mirror focusing an ambient energy, not twisting it. That claim is not tested here. It is queued as HI-12, “Does the mirror focus anything?” The twist results on this page say nothing for or against it, and twisted waves out of equilibrium support none of the episode's claims.
Sources: research/kozyrev-v5/RESULTS.md, RESULTS_v6.md and PRIOR_ART.md in the lab notebook (not public yet). Preregistered before the runs; self-graded, then challenged by an outside model (Codex), and the results record where that narrowed the claims.
The live instrument
The rig
Your records: measuredDashed: engine resultsPark the controls, let the reading settle, press Record measurement. Your point lands on these plots beside what the frequency-domain engine computed offline. Dashed slate is a guide, never data.
Reading the field
Waves leave the brass source and bounce off the spiral wall. The wall's distance from the centre grows with angle, so reflected waves leave with a twist: net angular momentum, measured on the dashed ring as the flux of angular momentum divided by the flux of energy. The spinning arrow is that reading; the small arrows are where the energy is actually flowing, averaged over a second or so.
Seal the box and the twist averages to zero. With nowhere to go, waves slosh both ways and the stored circulation hovers around nothing. Twisting needs escape. That is the time-reversal argument, run live.
Replace the speaker with heat: noise inside, a cold absorber outside. The radiation still twists. It is a tiny heat engine, like the published "Rytov pinwheels". Make the outside as hot as the inside and the net twist vanishes. That is why a spiral cannot harvest ambient fluctuations.
Drag the source behind the arm and the sign flips, as the v5 study found (D4). The mirror twin always gives the opposite sign exactly; that is symmetry, not physics. Hard walls (sound) keep the sign (v5 D2). Press Unstable step to watch the kind of bug that fooled the lab for eight months.
- 2-D, scalar and linear: sound in a flat layer or one polarisation of light, not full 3-D electromagnetism.
- The absorber is a smooth sponge, not a perfect PML. It reflects a little at low ω₀.
- The noise baths are band-limited around ω₀ (Q ≈ 3). The engine's exact bath is single-frequency, so compare them in sign and rough size, not decimals.
- Error bars are batch means over 16 stretches of the run. Very slow beats in a sealed box can still make them a little optimistic; give it time.
- The engine reference curves are for soft walls only. The unstable-step demo is the same class of bug as v2 (an explicit step past its stability limit), not v2's exact scheme.
- Nothing here supports Kozyrev's claims. The mechanism is textbook, and the equilibrium case gives zero.