The roar is the room
Hold a shell to your ear and it amplifies the sea, the saying goes. We put a spiral shell in a 2-D wave model of a room whose only sound is its own background noise. The shell does change what reaches the ear: some bands come out louder, others quieter. It adds no energy, and in a silent room there is nothing to hear.
The verdict
Nothing there
In a 2-D wave model, a shell reshapes the room's noise at the ear, up to about 1.9 times louder in some bands and a third quieter in others, but adds no energy: an ear at room temperature draws no net power from it, and a silent room stays silent. The preregistered check that the pitch follows the shell's size failed too, because its peak rule found no peak; a later overlay, not preregistered, finds that it does.
How the verdict was set. The run’s RESULTS proposed real effect, different reason: a band-shaped roar made of reshaped room noise, with a pitch set by the shell’s size. On 2026-10-18 the verdict was set from the lab’s verdict engine instead, at Ben’s instruction that its verdicts stand. The engine finds that the claim fails, that it is already known, and that the result is measured; in this wing that reads as nothing there. Two preregistered parts fail: no net gain, as expected, since this model cannot make any, and pitch follows size, because its peak rule found no peak.
With room noise as the only source, the energy at the ear reaches 1.88 times the no-shell value at the bottom of the sweep (k = 4.0), and is above it at 47 % of the 121 wavenumbers.
Between the shell’s resonances the energy at the ear falls to 0.65 times the no-shell value. The shell moves the room’s noise from one band to another.
An ear at room temperature draws no net power: at most 1.4×10−15 of what it exchanges, at five wavenumbers. With a powered source planted inside the shell the same probe reads 0.012 to 0.25.
With the room held at zero temperature, the energy at the ear is exactly 0. Whatever the shell plays is the room’s own noise.
A folk claim. Hold a shell to your ear and it amplifies the sound of the sea. No show or source is cited for it in the lab’s registry.
“Amplifies” can mean two things, with opposite answers. Louder in some bands, at the ear, is what a resonant cavity does to the noise around it. More power out than in is what nothing passive at room temperature can do. The episode tests both, and where the sound comes from.
Preregistered in PREREG.md (commit 728a7ed) before any code; Addendum 1 before any code; Addenda 2 and 3 after the run, declared as such, each before the check it governs. Self-graded, with no outside review. Lab files are listed at the foot of the page.
The setup
A rigid spiral wall, the golden log spiral the wing has used since episode 1, makes one turn from radius 0.1 to 0.68 inside a 2-D box. The box’s absorbing edge is held at room temperature, so the only sound in the room is its own thermal noise. The ear is a point just inside the shell’s mouth.
The measure is the total wave energy at the ear, divided by the energy at the same point with no shell, at 121 wavenumbers from k = 4 to 40. Averages over the noise are computed exactly, with no sampling. A second shell, scaled to 0.8 of the size, tests whether the pitch follows the size.
The checks
Every check below was written down before the code. Each planted fault had to turn its check red once; one could not be run.
| Check | Result | Numbers |
|---|---|---|
| The wave engine’s own tests and planted bugs | pass | 15 passed; all mutations caught |
| Noise beside a rigid straight wall, against 1 + J0(2kd) | pass | max error 0.048 / 0.025 / 0.016 at k = 10 / 20 / 30; limit 0.10 |
| … against the soft-wall formula instead | red, as it should be | 1.83 / 1.55 / 1.02 |
| Noise statistics exact at the ear | pass | 5.5×10−14 |
| Reshaping: at least 1.5× somewhere and at most 0.8× somewhere | pass | max 1.88× (k = 4.0), min 0.654× |
| No net power to an ear at room temperature | pass | at most 1.4×10−15 at k = 8, 14, 20, 26, 32 |
| … with a powered source inside the shell | red, as it should be | 1.2×10−2 to 0.25 |
| Pitch follows size, by the preregistered peak rule | missed | no peak of 2 or more inside k = 4 to 40 for the full shell; the 0.8 shell peaks at k = 5.30 |
| … with an unscaled shell compared as if scaled | could not run | the full shell had no peak to compare |
| Silent room | pass | exactly 0 |
| Finer grid, N = 361 against 241 | pass | 1.217 vs 1.200 (k = 20); 0.999 vs 1.009 (k = 30) |
The miss. The full shell’s strongest low peak, 1.88, sits on the bottom edge of the sweep at k = 4.0. The rule needed a peak of at least 2 inside the sweep, so it found none, and the planted fault for that check could not be run. The check stays failed. No gain could never have failed: it is an identity of this noise model, so it checks the instrument and says nothing new about real shells. The claim holding was out of reach by construction, and the preregistration said so before the run.
Two checks added after the run
Declared as post-hoc, each committed before it ran. Neither changes the preregistered result.
| Check | What it found | Numbers |
|---|---|---|
| A low-wavenumber sweep, k = 2 to 6 | both shells peak at the same place: a resonance of the simulated room, not of the shell | k ≈ 2.26 for both sizes |
| Overlay the two curves and fit the scale between them | the 0.8 shell’s curve is the full shell’s, shifted in pitch by its size | best scale 0.7975 for a true 0.80; correlation 0.96, against 0.08 unscaled |
So the pitch does follow the size: the smaller shell gives the same curve at a higher pitch. The preregistered test asked that question of the lowest peak, and the lowest strong peak fell outside the sweep.
What a listener hears, in this model, is room noise boosted in some bands and cut in others, with bigger shells giving lower-pitched roars. It is the same reshaping HI-12 measured for the spiral mirror.
- What a person perceives.
- A real 3-D shell, and air damping inside it.
- The blood-flow explanation. The model has no blood; its prediction is that in an anechoic room the shell’s roar disappears, and any roar left over comes from the body.
Prior art: UC Berkeley’s “Ask Marian” (2007) and WSU’s “Ask Dr. Universe” (2016) give the same ambient-noise explanation; Helmholtz-resonator acoustics is textbook; HI-12 in this wing measured the same reshaping. No novelty claimed.