Every post on this blog compares a phone to some professional instrument — a lux meter, a stroboscope, an octave-band analyser — and every one of those comparisons owes you the same disclosure. So here it is, once, in full: the map of where a phone genuinely competes with dedicated instruments, where it is an honest screening tool, and where it is simply the wrong device and pretending otherwise would be malpractice.
The map turns out to have a shape, and the shape follows from one piece of physics.
Frequency is nearly free. Amplitude is expensive.
Ask a phone when something happened, or how often, and it answers with the precision of a quartz crystal oscillator — the same technology that keeps wristwatches accurate to seconds per month. Timing is the thing consumer electronics does superbly, because everything inside the phone already depends on it. A frequency measurement is a counting problem: count cycles against a clock that's good to parts per million, and the sensor's quality barely enters. The microphone can be cheap, coloured and inconsistent — the pitch of a plucked bike spoke still lands where physics put it.
Ask a phone how much — how many pascals of sound pressure, how many lux, how many candela — and you've asked a completely different question. An absolute amplitude measurement is only as good as the chain of calibrations connecting your sensor to a national standards laboratory: reference instrument calibrates transfer standard calibrates your device, each link documented, each link decaying with time and temperature. Professional instruments are expensive largely because they carry that chain. A phone carries none of it. Its microphone sensitivity varies unit to unit; its camera's response depends on lens, sensor and firmware; none of it was ever measured against anything traceable.
Almost everything on this map follows from that asymmetry.
Tier one: where the phone genuinely competes
Some measurements a phone makes are not approximations of the professional version. They are the measurement.
- Frequency and timing. The pitch of a plucked spoke, the flicker rate of an LED, a resonance peak in a spectrum, a heart rate read off the fingertip. These are cycle-counting problems against a quartz clock, and a phone's answer is as trustworthy as a dedicated instrument's within its bandwidth.
- Ratios on the same sensor. The two-phone insulation test and the glazing transparency rig divide one reading by another from the same receiver, cancelling the microphone or camera entirely — the same design a professional would use if handed uncalibrated gear.
- Trends on one device. The same phone measuring the same washing machine monthly doesn't need to know its accelerometer's absolute gain; the drift between readings is real even when the readings' absolute scale isn't.
- Being there. The least glamorous advantage is decisive in practice: the best instrument for a decision made in a supermarket aisle is the one in your pocket. A measurement you actually take beats a better one you don't.
Tier two: honest screening
The middle tier is where most phone measurement lives, and where honesty matters most. A screening instrument answers "is this worth a closer look?" rather than "what is the certified value?" — and that's a genuinely useful question, as long as nobody upgrades the answer in the retelling.
There's encouraging evidence here, and it comes with a built-in caveat. When NIOSH researchers tested smartphone sound-measurement apps against reference instrumentation, the best performed within about 2 dB of the reference — genuinely respectable. The caveat: of nearly 200 apps examined, only a handful met the criteria to even be tested, and the spread across the rest was wide. The lesson isn't "phones are accurate" or "phones are junk"; it's that the phone's hardware admits a decent measurement if the software is honest about the pipeline, and nothing about the app store enforces that.
Screening-tier examples across the platform: band levels without the calibrated absolute scale of an IEC 61672 meter; illuminance without a lux meter's certified photopic response; the hearing screen, which measures its attenuation exactly but assumes its absolute anchor with a stated ±10 dB error bar; colour as comparison rather than colorimetry. In each case the tool's job is to be explicit about which half of the claim is measured and which half is assumed.
The screening contract: a screening result may only ever escalate — "this looks wrong, get it measured properly" — never certify. The moment a screening number is used to prove compliance, close a dispute or clear a safety requirement, it has been used outside its design, and no amount of app polish fixes that.
Tier three: the wrong instrument
Some jobs a phone cannot do, and the failure isn't a software gap — it's physics or law.
- Anything requiring traceable calibration. Occupational noise compliance wants a sound level meter meeting IEC 61672-1 class 1 or 2, calibrated before and after the survey. Workplace lighting audits want an illuminance meter characterised per ISO/CIE 19476. A phone can't meet these because the standards specify tolerances on properties nobody measured on your phone — and the certificate is the product.
- Legal and safety metrology. Noise disputes, lease conditions, machine-safety guarding, anything a court or regulator will read: the instrument's paperwork matters as much as its reading.
- Precision out of the sensor's league. A machinist's ±0.02 mm lives with calipers, not a camera. Sub-millisecond vibration analysis above a few hundred hertz outruns a phone accelerometer's sampling. Ultrasound, thermal imaging and NIR spectroscopy need sensors the phone doesn't carry — which is why no tool here claims to measure a fruit's sugar content: that's near-infrared physics, and a camera simply doesn't have the wavelengths.
- Diagnosis. The hearing and vision activities are screens that say "get this looked at", never clinical results. The pulse tool reports a rate, never a rhythm diagnosis.
The map in one table
| Measurement | Professional instrument | What the phone honestly does |
|---|---|---|
| Frequency, blink rate, RPM, pitch | Frequency counter, tachometer, stroboscope ($50–$2,000) | Competes. Quartz-clock timing; sensor quality barely matters. |
| Transmission loss, transparency (ratios) | Building-acoustics rig, spectrophotometer ($5,000+) | Competes on the ratio, with wider error bars and no certificate. |
| Sound level, band levels | Class 1/2 sound level meter ($300–$10,000, calibrated) | Screens. Relative levels and spectra solid; absolute dB assumed, not certified. |
| Illuminance | Certified lux meter ($100–$2,000, characterised response) | Screens. Good for ranking and trends; absolute lux is an estimate. |
| Colour | Colorimeter / spectrophotometer ($300–$20,000) | Screens. Same-conditions comparison, not colorimetric coordinates. |
| Dimensions | Calipers ($20–$200), laser distance meter ($50–$500) | Screens. Percent-level with a reference object; never tolerance work. |
| Hearing / vision thresholds | Audiometer in a booth; optometrist's lane | Screens, with stated error bars; escalates, never diagnoses. |
| Compliance, disputes, tolerances, spectroscopy | Calibrated, certified, traceable instruments | Not the right instrument. The phone should say so, and here it does. |
Why the honesty is load-bearing
It would be easy to blur these tiers — every measurement app's marketing does. This platform can't afford to, for a structural reason: every scan is designed to become a labelled training sample, and a dataset built by users who were misled about accuracy is a dataset of misplaced trust. The tools refuse readings they can't defend — the hearing test rules out bands the room is too loud for, the transparency rig discards captures that can't confirm the emitter's blink — precisely because a smaller true claim compounds and a larger false one doesn't.
So use the tiers as they are. Let the phone count, time, compare and trend — it's genuinely good at those. Let it screen, and act on the escalations. And when the job needs a certificate, hire the instrument that carries one. Knowing which question you're asking is most of measurement; the rest is in the tools overview.