The scan-mode narrative: what happens, step by step, when the machine looks at an object — first in plain words, then in the terms the engineering uses. Size-free — the physics privileges ratios, not dimensions.
The machine has no camera — no lens anywhere, nothing inside the chamber looks at anything. Instead, the walls sing to the object and listen to how it sings back.
You already know the trick. Tap a wine glass and the note tells you about the glass — shape, thickness, whether it is cracked. A doctor's ultrasound does it deliberately: send sound in, listen to the echoes, draw the baby. A CT scanner does it with rays from many angles, folding all the views into one solid picture. A hologram does it with light: record how waves bounced off a thing, and you can replay the recording to get the image back.
This machine is all three of those at once, turned inside out. Instead of a scanner that moves around a patient, the object sits still inside a closed room whose entire wall surface is the scanner — small speakers, microphones, and antennas everywhere. To scan, the walls take turns asking questions: a brief, quiet chirp of sound containing many pitches at once; a pulse of radio-band light. The waves wash over the object — some pass through, some bounce, some land exactly on a note the object likes to ring at, the wine-glass effect at many pitches simultaneously — and everything that comes back is recorded at every wall and folded into a picture of what must be sitting in the middle.
The operator's one-sentence framing is correct as stated: a scan is ultrasonic and electromagnetic holographic tomography — "tomography" meaning the CT-style many-angles-folded-into-a-volume part, "holographic" meaning the recordings keep the waves' timing, not just their loudness, so they can be replayed. And yes — among the electromagnetic probes are "flying doughnuts": doughnut-shaped pulses of light — real, laboratory-made, published measured physics, not a metaphor — which cross the chamber keeping their shape, report their own position as they fly, and come back carrying an unusually rich imprint of what they passed — though this machine using them as a routine scanning tool is still ours to demonstrate. The rest of this paper walks down from this picture into the machinery, one notch more technical per section.
One phrase from the sensing rulings carries through the whole paper: the machine sees matter in the terms by which it holds it. The same contrast that makes a grain visible is the contrast the built field grips it by — the scan is not an accessory bolted onto the builder, but the same machine listening instead of pushing.
None of the imaging physics here is exotic. Ultrasound computed tomography (USCT) — a ring or bowl of transducers surrounding tissue, transmission and reflection recorded on all sides, sound-speed and attenuation maps reconstructed — is MEASURED medical art, in clinical use for breast imaging; microwave tomography — the same geometry with EM antennas, reconstructing dielectric contrast — is likewise published, working art. Scan mode is these two instruments merged into one chamber and pointed at objects instead of patients.
Three structural — not clever — advantages make the chamber a better platform than a medical scanner:
Register: the tomography itself is MEASURED art; the chamber's structural advantages DERIVED from its geometry; the composed instrument — this chamber performing USCT-class imaging on arbitrary objects — design until the bench signs it (Principles §8b).
The sound side asks in coded chirps — brief multi-frequency
bursts at milliwatt level, low enough that nothing in the volume moves.
One gate speaks, all ports listen; then the next gate, or a coded
combination. The v1 schedule is one-gate-at-a-time; from milestone M5, a
Welch–Costas schedule — because 13 is prime with primitive root 2,
a perfect ambiguity-free hopping schedule exists for exactly 12 gates
(slot k → gate 2ᵏ mod 13) — lets all gates interrogate
simultaneously without colliding. Owner:
Field Compiler §15; the
acoustic elements (horns, meshes, drive laws) are
The Phonons.
The electromagnetic side speaks the machine's toroidal-pulse alphabet (ruling R2-b: the bore optical channel and the microwave face are one full TX/RX holographic alphabet — six decades of frequency, both directions). The "flying doughnut" is a single-cycle, space-time-nonseparable toroidal EM pulse — a laboratory object, not a design sketch:
The EM sweep also carries what sound cannot ask: dielectric response for material class, and optical spectroscopy down the bore for composition (Principles §5, Scan step 6). And the cross-carrier verbs add free viewpoints: light writes sound — a laser pulse absorbed on any visible surface, including the workpiece, launches ultrasound from that spot with nothing moving — and reads sound (schlieren/vibrometry). Owners: The Two Carriers §6b, Field Compiler §17.6.
Formally, a scan measures the object's scattering matrix S(f): for every frequency in the band, how a call from each gate maps to the response at every port. The pipeline, in summary (owner, with the full recipe: Field Compiler §15):
.pattern — the chord list, ordered by
signature contribution, calibration reference in the metadata.The chord list is the machine's native compression — an object stored by the resonances it rings at and the drive that addresses each one, sharpening progressively as chords arrive (the JPEG analogy is Principles §0–§1's). What makes this more than a file format is Principles §6: scan and build are one operation run in opposite directions — the same chord list is at once the scan's output, the builder's drive recipe, and the build's acceptance criterion, met when the workpiece rings true. Milliwatt listening is always on, so every instant of fabrication is also a scan; and the EM octaves ring up ~10⁴× faster than the acoustic chords they check, so verification is effectively continuous: acoustic builds, EM verifies.
Turning port records into a picture of the interior is the hard half of scan mode, with the sharpest failure modes. The full treatment is software-owned (Field Compiler §16.3); this is the narrative and the discipline.
The reconstruction ladder runs four rungs, cheapest first — separately selectable operators over the same records, so outputs can be compared: L0 delay-and-sum backprojection (milliseconds; a reflectivity image — a 3D B-scan, not a geometry, and the UI must say so); L1 DORT — the SVD of S(f) taken directly, each significant singular vector one scatterer, with a free reciprocity check that flags calibration faults; L2 regularized linear inversion against the fast forward model; L3 full-waveform inversion — honest about the fact that recovering an object, rather than a field, from boundary data is nonlinear inverse scattering.
The diversity budget. No amount of field control computes its way out of a measurement gap. Transmit-pattern cleverness — beamforming, coded excitation, holographic drive — adds no independent information: every pattern is a linear combination of the element responses, buying signal-to-noise and conditioning, never new k-space; adaptive "smart scanning" allocates the budget, never enlarges it. Only two axes add genuinely new information: frequency (each frequency probes new k-space, and the printed holograms steer by frequency, so every tone is a different illumination geometry from a static aperture) and position, real or virtual — wall-mirror virtual sources, the instance's rotating door, and §2b's photoacoustic sources, an acoustic source anywhere light can see with nothing moving.
Coverage — rendering what the machine does not know. Some facets are unmeasurable from the aperture as posed (the specular gap): that is a null-space of the measurement, not an algorithm failure, and no regularizer recovers it. So every voxel of a reconstruction carries a tri-state that is never collapsed and never interpolated: OBSERVED-OCCUPIED, OBSERVED-EMPTY, NEVER-OBSERVED — and confirmed-empty and never-observed must never blur into one another, because a filter that averages them manufactures a plausible surface out of pure ignorance.
Validation without ground truth. With no turntable (the design aims at zero moving parts), correctness is checked by held-out prediction — reconstruct from a subset of gates or frequencies, then predict the withheld measurements; a wrong cavity model cannot predict data it was never fitted to — plus reciprocity, inter-rung agreement, and, for absolute accuracy, a machined reference phantom of known geometry (Field Compiler §18).
Ruling R12 fixes what the machine may honestly claim, rung by rung — quoted from the photon paper with its register tags intact:
| rung | claim | register |
|---|---|---|
| compound broadband full-waveform tomography, in-air | surface features at λ_min/2 of the working band — the impedance cliff makes solids-in-air opaque, so this is scattering tomography of surfaces: silhouettes, shadows, multi-bounce, fully invertible (instance: ≈0.85 mm at the plan-A air band — hardware spec). Interior κ/ρ/α phase maps require the fluid-coupled roadmap rung | DERIVED on MEASURED art (USCT) |
| band extrapolation | ~2× more, log-SNR-bounded | DERIVED |
| harmonic imaging | a few × more | MEASURED art |
| resonant near-field structure + broadband decoding | λ/25–λ/30 class | MEASURED elsewhere (Lerosey & Fink, Science 2007); in-chamber OPEN |
| discrete-scatterer localization + photoacoustic beacons | λ/100-class positions — the payload is discrete and carries its own beacons, so localization, not classical resolution, is the machine's working question | MEASURED art (ULM: Errico et al., Nature 2015) |
The last row is the one that matters for building: localizing known discrete scatterers is far easier than classical imaging, with measured art (ultrasound localization microscopy) two orders below the wavelength.
TX/RX conjugacy — the identity that closes the loop. A phase-kept scan is hologram acquisition; the same recording, replayed conjugate, is a trap around the thing imaged (time reversal). And the same Δκ/Δρ contrast vector appears in Born scattering (what the machine senses) and in the Gor'kov force (what it grips). The machine sees matter in the terms by which it holds it — §0's slogan, now as an identity.
The product of a scan is a .pattern file: the chord list
of §3, ordered by contribution, streamed progressively, calibration
reference in the metadata
(Memory &
Compute owns the format; Field Compiler §19 the encoding). Because the
scan output is also the build acceptance criterion, the file carries an
honesty obligation no ordinary scanner file has:
provenance: measured | inferred tag, and so does every
reconstruction derived from it. Inferred content — anything a prior or
regularizer filled in — is rendered visually distinct, never merely
footnoted, and is excluded by default from the build acceptance
criterion: you may look at it, you may not fabricate to it, and promoting
an inferred chord to load-bearing is an explicit, logged user action.
Otherwise an inferred interior becomes a fabricated-from-a-guess interior,
laundered through a fabrication step that looks like verification.
(Owner: Field
Compiler §16.3.)Alongside the chords, the coverage tri-state of §4 ships with every reconstruction — export, receipts, and gate reports all carry the measured/inferred split and coverage statistics. Detail-adaptive chord budgeting is first-class: spend chords where the object rings complex, stop early where it is homogeneous — resolution in this format is chord count.
Scanner-first — the staging note. The hardware spec stages the product so that the first build, powered below the assemble/dissolve thresholds, ships as a complete desktop 3D scanner at milliwatt class — no passive camera; appearance is computed from composition. First staging rung: acoustic (exterior geometry plus eigenmode fingerprint) plus active optical; second: the EM materials suite (dielectric spectroscopy, eddy-current σ/µ, Raman/hyperspectral). Scan mode is the first product the architecture ships; the copy loop (scan → pattern → viewport → build, when powered) runs end-to-end from day one (hardware spec, product-staging row).
Claims are tagged MEASURED (published or bench-verified art), DERIVED (arithmetic on measured quantities), or OPEN (registered bet with named falsification). This paper narrates; it introduces no new results and upgrades nothing it quotes.
.pattern is the machine's claim, not yet
demonstrated.Where everything lives: probe elements — The Photons / The Phonons; pipeline and reconstruction — Field Compiler §15–§18; the mode's place among the three — Principles §5–§6; what the first scanner ships as — Hardware Spec. The sibling narratives are Assemble and Dissolve.