Cover your phone's camera and torch with a fingertip and the screen goes a deep, glowing red. That red is torch light that has travelled through your finger — through skin, tissue and blood — before reaching the sensor. And it is not constant. With every heartbeat, a little more blood is pushed into the fingertip's vessels; blood absorbs light; so with every heartbeat, slightly less light gets through. The glow dims and recovers, dims and recovers, in time with your pulse.

That is photoplethysmography — PPG — the same principle behind the clip a nurse puts on your finger and the green LEDs on the back of a smartwatch. "Photo" for light, "plethysmography" for measuring changes in volume. The Pulse tool does the transmission version: torch on one side of the tissue, camera on the other, and the signal is the rhythmic modulation of what passes between them. It reports two things: a rate in beats per minute, and how strong the pulsatile signal actually was. It never reports a diagnosis.

Why the tiny signal is recoverable at all

The pulse-driven change in transmitted light is small — a modulation of a few percent, often less, riding on a large steady level. Recovering it depends on everything else holding still, and that is where the platform's usual capture discipline does the real work.

First, the torch. It is the only light source that matters here, and it is a known one: fixed position against the finger, fixed output, on for the whole capture. The illumination side of the measurement is a constant, so any periodic variation in what the sensor receives must have been imposed by the medium in between. Ambient light barely enters — the fingertip pressed over the lens seals most of it out, and what leaks in is steady rather than periodic.

Second, the camera stops adapting. A phone camera's default behaviour is to fight exactly the signal we want: auto-exposure sees a slightly dimming frame and nudges gain up to compensate; auto-white-balance re-tints frames as it second-guesses the red flood. Either one overwrites a few-percent physiological modulation with a few-percent software correction. So the Pulse tool locks AE and AWB for the duration of the capture — the frames are measured on one fixed scale, and the only thing left moving is the light itself. It is the same rule that makes the reflectance and translucency tools work, and the optical cousin of the tap test's unprocessed-audio path: switch off the helpfulness, keep the physics.

From there the analysis is a matter of averaging the frames down to a brightness-over-time trace, finding the dominant periodicity in the plausible range, and — just as important — measuring how much of the trace's energy that periodicity actually explains. That last number is the signal-strength indication reported alongside the rate. A firm, still fingertip gives a strong, clean oscillation; a trembling grip, a cold finger or a half-covering press gives a weak one. The tool tells you which you got, so a low-confidence reading can be discarded rather than believed. Everything runs on the device; no frame leaves the phone.

This is not a medical device. The Pulse tool reports a rate and a signal strength — never a diagnosis, never health advice. It is not cleared, certified or validated as a medical instrument, and it should never be used to make a health decision. If a reading matters to your health — a rate that seems too high, too low, or irregular — that question belongs with a clinician and proper equipment, not with a phone camera.

The physics doesn't know it's a finger

Here is the part that makes Pulse more than a novelty heart-rate app: nothing in the measurement is specific to fingers. The tool measures periodic modulation of transmitted light by a medium pressed against the lens. A fingertip qualifies because blood volume pulses. But so does anything else that pulses.

Press a length of soft tubing from a small pump against the lens and torch, and every pump cycle changes the liquid volume — and therefore the light absorbed — in exactly the way a heartbeat does. The tool reads the pump's cycle rate the same way it reads a pulse. Some places this turns out to be genuinely useful:

This generality isn't a stretch of the tool; it's a property of measuring the medium rather than assuming the subject. The rate reading is agnostic about what produced the rhythm — which is precisely why, in the medical direction, it can only ever be a number, and in the plumbing direction, it's a perfectly good instrument.

What a rate is good for, kept honest

Because the output is a rate plus a confidence, the honest uses are the ones a rate actually supports. Checking your resting rate out of curiosity, before and after a run. Verifying that a pump is cycling, and how fast. Watching a machine's rhythm drift over weeks. Each reading can be saved under the object it belongs to — a person is not a catalog object, but a pump is, and for the curiosity readings an unassigned capture works fine — and each can take a ground-truth label later, the same as every other tool: the pump that was indeed failing, the line that was indeed clogged. That is the dataset flywheel, applied to plumbing.

What the readings are not good for is anything diagnostic, and the tool's design keeps saying so: it reports beats per minute and signal strength, and stops there. No zones, no warnings, no wellness scores. The restraint is the feature. A number you can trust for what it is beats a verdict you can't.

One trick, many instruments

Pulse belongs to a small family of tools that pair the torch with the AE/AWB-locked camera to form a controlled optical instrument — its siblings measure surface reflection and translucency with the same light and the same locked detector, pointed at different questions. And that family is itself one branch of the larger project of treating a phone's sensors as instruments rather than conveniences. For the map of the whole territory, start with the tools overview.