Somewhere in a second-hand shop, you are holding a bowl. The label says brass. Your thumbnail says maybe. The dealer says definitely. What would it take — short of a laboratory — to say something defensible?
The physics is on your side, and it's older than the shop. Tap the bowl and it rings; how that ring decays is set largely by the material's internal friction, not by the bowl's shape. Metals ring long and clean. Glass and ceramic ring bright with fast-fading overtones. Wood damps politely; plastics smother the ring almost before it starts. Aircraft inspectors have exploited exactly this for decades as the "coin-tap" test for hidden delamination, and its mechanics are properly published science — Cawley & Adams' 1988 analysis in the Journal of Sound and Vibration is the classic treatment. The single-capture version of this idea is Sentio's Tap test.
But one tap is one opinion, and this post is about the machinery that turns opinions into a verdict: the material-identification activity, the platform's clearest example of what a guided activity is for. Where a tool is a capture, an activity is a protocol — a chain of tools, each covering the others' blind spots, with rules about when the chain is allowed to conclude.
Step one: tap, and tap again, and again
The activity opens by demanding at least three usable taps. Each tap is a full ring-down capture and gets its own classification — a score across six material families, with an honest confidence attached. And each per-tap classification is allowed to decline: a clipped recording, a tap with only one detectable resonant mode, or a noisy room all shrink the confidence, and below a floor the tap simply casts no vote rather than a shaky one.
Then comes the rule that gives the activity its character: at least two of three taps must agree on the family, or there is no verdict. Not a lower-confidence verdict — none. Disagreement between taps on the same object isn't noise to average away; it's information that the procedure wobbled: different spot, different striker, a grip that damped the second tap. The right response is another pass with a steadier hand, and that's exactly what the activity asks for.
Step two: the probe — a second sense as a veto
Next, optionally, the activity switches sensors entirely. The phone is pressed against the object and the haptic-response tool takes over: the vibration motor buzzes the object with a known excitation while the accelerometer and microphone record how it answers — including how fast the response dies when the buzzing stops.
Here's the interesting design decision: the probe's evidence is coarser than the tap's — essentially "this rings" versus "this damps" — so the activity never lets it pick the family. It lets it veto. If the taps say METAL and the probe's ring-down is over in milliseconds, something is wrong — perhaps the taps caught a metal fitting on a plastic body — and the activity returns no verdict rather than a compromised one. If tap and probe agree, confidence gets a modest boost, capped well below certainty. Two independent senses agreeing is worth more than either alone; two senses disagreeing is worth exactly nothing, and the activity prices both correctly.
Step three: the reference — borrowing certainty
The final optional step is the cleverest, and it's pure protocol rather than new physics: tap something you already know. The known stainless pot, the known glass jar, the known oak board. If the mystery object's ring-down features sit close enough to the reference's, the verdict snaps to the reference's material — a comparison against ground truth you supplied, which is a far stronger claim than a classification against priors.
The comparison is careful in one crucial way: it matches on damping character — how sharply tuned the resonances are, how the highs persist, where the spectral weight sits — and deliberately ignores pitch and mode count, because those encode the object's size and shape, not its substance. A small brass bell and a big brass bowl disagree completely in frequency and agree closely in decay. Matching on the decay side is what lets your teaspoon vouch for a tray.
Families, not alloys — and one verdict that isn't a material
The activity names six families: metal, glass/ceramic, wood, hard plastic, soft polymer — and a sixth called damped-or-loaded, which deserves a sentence of appreciation. It's the verdict for objects whose ring says "something is absorbing energy in a way no clean material does": a cracked plate, a filled or laminated panel, a thing held too tightly. It isn't a material; it's a condition, and giving it a name keeps it from contaminating the real families. A cracked ceramic mug taps like neither ceramic nor plastic — the honest output is "damped: check for a crack or a laminate," not a coin-flip between wrong answers.
Families are also where the honesty about resolution lives. Damping cannot tell steel from aluminium — their internal friction is too similar — so the activity doesn't try. Glass versus fired ceramic, likewise. What a tap can support is the family; what it can't, the activity declines to invent. And there's a disclosure behind the disclosure: today's family boundaries are physics-informed priors, not a model trained on a large labelled tap corpus. That corpus is precisely what the activity is accumulating — every tap saves its full ring-down capture, and a later ground-truth label ("it was brass; the dealer was right") turns the whole session into training data. The classifier is designed as a socket the trained model drops into once enough people have tapped enough labelled things — the same flywheel as the fruit test, aimed at flea markets instead of fruit stands.
What the professionals use
| Professional instrument | Phone activity | Where the gap bites |
|---|---|---|
| Handheld XRF analyzer ($15,000–$50,000): reads elemental composition off the surface in seconds | Acoustic family from ring-down; probe and reference cross-checks | Alloy identity, precious-metal verification, regulatory scrap sorting — elemental questions need elemental physics |
| FTIR / Raman spectrometer ($10,000+): identifies specific polymers by molecular signature | "Hard plastic" vs. "soft polymer" as families | Which plastic — recycling QA, restoration, forensics. Damping can't see molecules |
| NDT tap-test instruments ($1,000–$10,000): instrumented hammers and automated tappers for aerospace composites | The same physical principle (Cawley & Adams), consumer striker, agreement protocol instead of calibrated impulse | Certified inspection of safety-critical parts — mapping a wing skin is a job for the calibrated tapper and a signed report |
| Density measurement (scale + displacement, ~$50 of kit): classic and surprisingly decisive | Non-destructive, no immersion, works on furniture-sized objects | When you can weigh and dunk the object, Archimedes still competes admirably with everyone above |
The pattern to notice — because it's the pattern of every activity on the platform: multiple captures before any conclusion; independent evidence used to veto, not to embellish; a reference measurement outranking any classifier; and "no verdict" as a first-class result. The hearing screen refuses untestable bands, the transparency rig refuses unconfirmed blinks, and this activity refuses disagreeing taps. A guided activity is precisely the codification of when not to answer.
For the raw physics of the single tap, read the materials tap-test post; for how the phone-shaking probe works, haptic response; and for the wider map of phone-versus-instrument trade-offs, the honest accounting.