Most measurement apps are a single instrument with a screen: a sound level meter, a compass, a light meter. You open it, you read a number, you close it. The number is gone the moment the app is.

Sentio is built the other way around. It assumes the number is worth keeping, that it belongs to something in your life, and that its meaning usually comes from comparison rather than from the reading itself. That assumption produces three concepts — tools, objects and activities — and once you have those three, the whole app is explained.

Tools: one unit of measurement

A tool is one way of asking the physical world a question with the hardware in your pocket. Tap a fruit and record the ring-down. Hold up the camera and read the color of a wall. Point the torch at a surface and measure how much light comes back. Play a sweep through the speaker and listen to what the room does to it.

There are twenty-six of them today, and they fall into four families that get progressively more demanding of the world around you:

That last requirement is a first-class property of every tool rather than a footnote in its instructions, so the Tools tab can filter on it: show me what I can do right now, alone, with nothing but this phone — or show me everything, because I have a second phone on the table.

A worked example: is this countertop solid?

You are in a shop, in front of a counter that is described as stone. Open the tap test for materials, hold the phone about a hand's width from the surface, and knock it firmly with a knuckle three times. Each knock takes about fifty milliseconds to say what it has to say. The app finds the onset of each one, isolates the ring-down, and reports two numbers: the dominant resonance frequency and how quickly the ring decays.

Solid stone rings high and dies slowly. A laminate over chipboard rings lower and dies fast. Hollow drywall barely rings at all. You are not reading a verdict off a screen so much as reading the difference between the counter in front of you and the one three aisles over — which takes another twenty seconds and is the comparison that actually decides the question.

A second example: is the fridge getting worse?

Your fridge has started making a noise. Open the sound spectrum tool, hold the phone a foot from the compressor, record five seconds. You get the frequency fingerprint: where the energy sits, which harmonics stand out, how loud the peaks are relative to the noise floor.

On its own that is a picture, not a diagnosis. Its value is that you can record the same five seconds next month, from the same place, and lay the two on top of each other. A new peak at three times the compressor's running frequency that was not there in August is a real finding about your fridge. An absolute number in decibels is not.

Objects: what the measurement is about

This is the part that makes the fridge example possible, and it is the concept most measurement apps skip.

Sentio keeps a catalog of the things you measure, nested four levels deep: a category (Appliances) holds an object (the kitchen fridge), which holds a scenario — a use case, like "compressor noise, front left, 30 cm" — which holds the measurements themselves. A scenario carries the tool it defaults to, so repeating a reading a month later is one tap, and it repeats the same setup rather than an approximation of it.

You do not have to file anything up front. A capture can be saved unassigned and attached to an object later, which is what usually happens in a shop: measure first, decide what it was afterwards.

Every measurement can also receive a ground-truth label after the fact. You tapped the melon on Tuesday; you cut it open on Saturday; the label goes on the Tuesday measurement, where it turns a recording into a training example from a real fruit on a real phone. That loop is deliberate: the classifiers get better because people used the tools, not before.

Why the catalog is the honest part of the product: a phone cannot tell you a countertop is granite, and any app claiming otherwise is guessing at your expense. What a phone genuinely does well is tell you that this surface differs from that one, or that this machine differs from its own past self. Comparison needs a place to keep the past — that place is the object.

Activities: several tools, one job

Some questions do not fit inside a single measurement. "Is my bedroom wall thick enough to block my neighbour's television?" needs a reference capture and a transmitted capture. "Should I get my hearing checked?" needs a sequence of tones at controlled levels, in the right order, with the results held together. "What material is this?" is more trustworthy when three taps agree than when one tap is confident.

An activity is a guided procedure that chains tools toward one job. It walks you through the steps, tells you what has to be true before each capture, and — this is the part worth insisting on — refuses to produce an answer when the captures disagree with each other. An activity that shrugs and averages contradictory data is worse than no activity, because it launders uncertainty into a number that looks confident.

Activity sessions are ephemeral by design. You run one, you get your answer, and nothing is written to the catalog until you press "Save to Objects" — at which point it writes ordinary measurements, filed under ordinary objects, indistinguishable from ones you took by hand. There is no second, parallel kind of data.

There are four activities today. There were seven: three of them turned out to be protocols that capture twice with one sensor, and a procedure that chains nothing is a tool wearing a costume, so they became tools instead. The concepts are meant to earn their place.

A platform, not a fixed instrument

The tool count in the second paragraph of this section has been wrong three times while this post was being written, and that is the point.

A measurement is stored as a tool identifier plus one opaque payload of features. Adding a new tool means adding an entry to a registry and the analysis that fills that payload — it is never a change to how measurements are stored, never a migration, never a reason for your existing history to become unreadable. That is a deliberate architectural bet, and it is what makes "the sensor list keeps growing" a promise rather than a marketing line.

So the honest description of the app is not "a light meter and a sound meter in one download". It is a platform for measuring physical values with the sensors you already own, storing them against the things they describe, and watching them move over time. New tools arrive; your fridge recordings from last year still line up next to this year's. The instrument gets better while the history stays intact.

And all of it runs on the device. Analysis takes under a second, the network is optional, and nothing leaves the phone unless you sign in and press sync. A measurement platform that stops working in a basement, a warehouse or a field is not measuring the world you actually live in.

Where to start

If you want the map of what the sensors can and cannot do, read what a phone can actually measure, and then the honest accounting against real instruments. If you would rather just try one, the web app records a real sound spectrum in your browser with no account at all — point it at something that hums.