Current compact optical methods

The original handle exterior and its two-camera, two-light packing are carried over unchanged from the earlier compact study. Only the tray mirror positions and orientations were searched again; the R24 6 × 6 mm prescription, the trays, the seat candidates, the devices and the mirror counts are those of the symmetric design these layouts replace.

The mirrors were searched against 7 patients’ arches at once (6 on the lower, where one arch is longer than the tray): the reference scan behind every earlier design and 6 further intraoral scans, each seated in the unchanged trays by a rule that reproduces the reference seats within 0.25 mm. A cap must clear every arch. Each candidate was traced at every shot of each arch's capture sequence, and a surface point counts at the best shot that reaches it.

Candidates were scored by a balanced objective: the sum over (arch, surface class) of log(C + 0.05), where C is the area-weighted value of that class on that arch and a triangulating pair is worth 0.3 + 0.7 × q_camera, with q = min(1, (target/pitch)^weight), target 0.15 mm per pixel and weight 2. The superseded search ranked min(q_camera, q_light), which the light head's raster reduced to the light leg alone. Mirrors are chosen as exact mirror-image pairs about x = 0, greedily and then by exchanging whole pairs while any exchange raised the objective. This is the best found bounded search, not proof of a global optical optimum.

Independent held-out tests on the reference arch use the same frozen sets as the earlier study: 8,000 upper samples (5,900 primary) with seed 20260920101 and 6,000 lower primary samples with seed 20260920102. Sampling is uniform by area.

The lower layout was searched and evaluated with an opaque tongue-and-plate proxy appended to the anatomy; the upper was not. Separate audits check complete body/wiring enclosure, intersections, remaining mirror seats, connected material, finite reflection paths, exported masks and intervals. Browser mesh decoding and STL generation are compared with the frozen physical files.

Exactly two cameras, two combined light heads, one shared driver strip and opaque wiring segments per layout. The tray contains mirrors and passive supports only.

Camera assembly reserves are 5 × 5 × 8 mm. Light-head reserves are 8 × 8 × 12 mm. The shared driver strip reserve is 12 × 40 × 4 mm. These are proposed assemblies, not qualified purchased systems.

Each camera uses a 4208 × 3120 ideal rectilinear surrogate with a 70° horizontal and 54.87° vertical field (fx 3004.8 px). Actual intrinsics and lens distortion have not been calibrated.

The structured source uses a point pupil and continuous rectangular cone of 109° horizontally and 95° vertically, based on the Himax TL25-W 940 nm fixed-dot projector. Virtual source coordinates encode the cone; they are not a physical pixel array.

Both source and camera routes are direct or use one finite reflection. The acute camera–source ray angle must be at least 10°. Full scan triangles, remaining housing, component and wire boxes, other mirrors and their backing surfaces obstruct routes.

White, 450 nm visible-blue fluorescence for caries and plaque/biofilm, and 850 nm class NIR flood are retained as separate module requirements. No UV channel is proposed. Their individual pupils, signals and diagnostic performance are not inferred from structured-source coverage.

MULTI-SHOT PRESCRIPTION. Every single-pose figure is ONE seated pose. The tray is captured over several rigid re-seatings and stitched; the stitched figures are in the coverage viewer and on the Specifications page, and cohort-study/README.md gives them for every arch of the search cohort. They are simulation under assumed priors, not measurement.

What the published percentages mean

Camera-visible and illuminated are separate fractions. Triangulation geometry requires both routes, each direct or with one finite specular reflection, and a camera–projector acute line angle of at least 10° (10°–170° directed angle). The surface normal must face the arriving ray. Full scan triangles, the revised opaque tray/handle, other mirror surfaces and component bodies can obstruct each segment. The four channels share a provisional optical aperture; identical surface access does not imply identical signal or diagnostic performance.

The independent held-out samples are uniform by surface area. Counts therefore estimate unique surface area, rather than the number of optical paths. The main denominator includes tray-facing and inner/outer side-facing regions. Superior-facing remainder is excluded; the upper palatal proxy is reported separately. Gums are present in these regions, but no validated tooth/gingiva separation exists. The recessed-surface proxy overlaps other rows and is not a clinical interdental segmentation.

Mirror search uses a finite set of measured mount seeds and a bounded set of apertures, orientations and curvatures. An optimistic shortlist is followed by evaluation against actual modified housing. Final support checks and held-out evaluation are distinct from this shortlist. The selected design is the best found by that search, not proof of maximum possible coverage.

Geometry and fit checks

The scan scale is 1.0 throughout. Large is the best found fit for each provided scan. Selected seated poses were checked in both directions for mesh-edge/triangle crossings and on 6,000 surface samples for clearance and embedded points. This does not prove that unscanned anatomy, coplanar contact, manufacturing tolerances or clinical insertion are acceptable.

Handle enlargement is explicit geometry. Camera, module and driver boxes are tested against original and revised envelopes. Cavities and finite optical corridors are actual Boolean subtractions. The residual material remains an optical obstruction. Mirrors use supported edge or floor locations, with a second geometric seating check against the actual material remaining after optical cuts and mirror pockets. Explicit passive bridges are included where needed. Socket proximity does not establish retention strength, tolerances or attachment design. A watertight shell is not a manufacturing certificate: remaining wall thickness, assembly access, thermal dissipation and sealing still need engineering review.

Lossless geometry storage

The full calculation meshes use shared triangle pools. Decoding preserves coordinates, winding and original face order; anatomy retains float64 precision and CAD retains its exact float32 STL coordinates. Downloaded STLs reconstruct these same checked triangles. This storage change does not alter the coverage calculations.

Independent checks

The downloadable validation records exercise known reflection solutions, coated-side rejection, reciprocity, finite-aperture misses, curved-surface intersections, opaque housing and opened ports. Independent box-overlap and mesh-query checks use separate numerical formulations. Passing these checks validates the tested calculations; it is not clinical or hardware validation.