The torch is the instrument: reflectance and translucency

A phone camera pointed at a surface tells you almost nothing quantitative about that surface, because the picture is dominated by whatever light happens to be falling on it. A kitchen counter photographed at noon and at dusk is two different images of the same material. The camera is a fine detector; the problem is that the illumination is unknown.

The phone, however, carries its own light source, a few millimetres from the lens. The torch isn't adjustable, it isn't colour-tunable, and it isn't calibrated in any laboratory sense — but it is consistent: the same LED, the same position, the same output every time you fire it. A known light plus a detector is the skeleton of a real optical instrument. Two of Sentio's tools are built on exactly that skeleton, and they point it in opposite directions: one bounces the light off a surface, one shines it through.

Reflection: the frame pair is the trick

The Reflection tool doesn't take a photo of a surface. It takes two: one with the torch off, one with the torch on, from the same position, and works on the difference.

That pairing is the whole idea. The ambient frame records everything the room contributes — window light, ceiling lamps, shadows. The torch-lit frame records the same scene plus one thing more: the surface's response to the torch. Subtract the first from the second and the room mostly cancels out. What remains is how the surface answered a light whose intensity, position and spectrum are the same on every run. You've replaced "a photo under whatever light" with "this surface's response to a known stimulus" — and that quantity is comparable from one measurement to the next.

For this to work the camera has to stop being helpful. Auto-exposure would see the torch-lit frame as brighter and quietly turn the gain down; auto-white-balance would re-tint each frame independently. Either one destroys the subtraction, because the two frames would no longer be measured on the same scale. So the tool locks AE and AWB for the pair — the same discipline, for the same reason, that the tap test uses the phone's unprocessed audio path. The pipelines phones ship with are built to make things look and sound good; measurement needs them out of the way.

What is this good for? Anywhere the interesting property is how a surface returns light rather than what colour it is:

• Comparing finishes — two paint samples, two countertop offcuts, satin vs semi-gloss, measured instead of eyeballed under showroom lighting.
• Gloss vs matte — a glossy surface throws the torch back sharply; a matte one scatters it. The difference frame separates the two far more cleanly than a single photo can.
• Wear on a coating — measure the same spot on a varnished tabletop or a car panel over months. As the coating dulls, its response to the same light drifts, and because each reading files under the same object in your catalog, the drift is a history you can actually see.
• Did the cleaning work? — before-and-after readings of the same spot answer the question with a number rather than a squint.

Translucency: candling, with the lens as the far side

The sibling tool flips the geometry. Instead of bouncing torch light off a surface, you press the sample against the camera lens with the torch behind it, and the tool measures how much light makes it through — and, usefully, how that transmission differs between the red, green and blue channels. This is candling — the old trick of holding an egg to a lamp — with the phone supplying both the lamp and the eye.

The colour-channel split matters because materials rarely attenuate light evenly across the spectrum. A leaf passes red more readily than blue; an egg's contents shift the transmitted colour as the egg develops; fruit skin transmits differently as pigments change. So the reading isn't just "how much got through" but "how much of each colour got through" — and the ratio between channels is often more informative, and more stable, than any single level.

Use cases where "how much light passes, and what colour survives" is the question:

• Egg candling — the classic, now with numbers you can compare across a batch.
• Leaf health — the same leaf position on the same plant, tracked week to week.
• Film and paper opacity — comparing tracing papers, diffusion gels, packaging films from the same roll or between candidates for a job.
• Fruit skin — pigment and skin changes show up in what transmits, a different window onto ripening than the acoustic one.

The honest caveat: same setup, or no comparison

Translucency readings are comparable between runs of the same setup, not between setups — and it's worth being plain about why, because the reason is physics, not a software limitation.

When you press a sample against the lens, the geometry of that contact becomes part of the measurement. How hard you press changes the sample's thickness and how flush it sits. Where exactly it sits changes the path length from torch to sensor. An air gap scatters light; a tighter contact doesn't. None of that is recorded, so none of it can be corrected for. Two readings of the same leaf, held the same way on the same phone, are a fair comparison. A reading from your phone against a reading from a friend's — different torch position, different lens, different pressure — is not, and the tool won't pretend otherwise.

This is the same reasoning, applied honestly, that shapes everything else in the platform: use readings relatively, within a consistent rig, and they're trustworthy; treat them as absolute universal values and they're not. The two-phone sound-insulation activity goes to some length to build that ratio structure into its own protocol, for exactly this reason.

Two tools, one instrument

Reflection and Translucency are the same instrument in two configurations: a fixed light, a locked detector, and a feature computed from a comparison rather than a single frame. Both run entirely on the phone, both save their readings into the objects catalog so a surface or a sample accumulates a history, and both can take a ground-truth label after the fact — the coating that did fail, the egg that was fertile — so every measurement is a future training sample.

They're two entries in a longer list. For the full tour of what a phone's sensors can be turned into, start with the tools overview.