You're comparing window films at the hardware store. You're deciding whether the "low-e" glazing quote is worth it. You're holding two pairs of sunglasses, both labelled polarized, one suspiciously cheap. In each case the question is optical and simple — how much light does this thing pass, and does it care which way that light vibrates? — and in each case the honest answer is a ratio, which is exactly what two uncalibrated phones are good at.

The transparency rig

One phone (or a browser) is the emitter: it blinks its screen at a fixed 2 Hz against one side of the pane. The other phone is the measurer: its camera watches for that blink and records the luminance swing — how strongly the brightness pulses at exactly the emitter's rate. Two captures:

  1. Direct — the two phones face each other with nothing between them, same geometry.
  2. Through — same geometry again, with the pane in the path.

The pane's relative transparency is the through/direct swing ratio. Everything that isn't the pane — the emitting screen's brightness, the camera's sensitivity, the phone models involved — appears identically in both captures and divides out. This is the same-receiver-ratio design that powers the two-phone sound insulation test, translated from a speaker and microphone to a screen and camera.

The blink rate is what makes it work in a lit room. Ambient light doesn't pulse at exactly 2 Hz, and auto-exposure chases the mean level, not the modulation — so the swing at the known rate isolates the emitter's contribution from everything else in the scene. If a capture can't confirm the blink rate, the activity refuses to compute a ratio from it: a swing that isn't provably the emitter's is a retryable capture problem the app names, never a "zero transparency" finding.

One rule is enforced with deliberate stubbornness: both captures must read the same patch of the frame. The capture region — one pane of a window rather than the whole scene — is chosen before measuring, and the moment the first capture lands it locks. Re-aiming means discarding every capture and starting over. That sounds strict until you consider the alternative: silently pairing readings of two different patches and calling their ratio a measurement.

Where one number stops being enough

A tinted, coated or low-e pane isn't equally transparent at every colour — those coatings typically block blue far harder than red, and an overall luminance ratio averages exactly that away. So the activity also splits the transmission across the camera's red, green and blue channels, applying the same discipline per channel: a channel the emitter barely drove reports nothing rather than a ratio of two noise figures. The most robust figure is each channel's transmission relative to the overall one — a ratio of ratios, so even an exposure shift between the two captures divides out of it. That per-colour selectivity is what actually distinguishes "grey tint" from "heat-rejecting coating" on two panes that look similarly dark.

To be precise about the claim: camera luminance is gamma-processed, so the result is a relative optical transmission of this rig — a solid basis for comparing panes and films against each other, not a calibrated laboratory %T or haze number. The activity doesn't pretend otherwise.

The quarter turn

Here's the elegant part. A phone's screen emits polarized light — a side effect of how LCDs and OLEDs are built. That accident turns the same rig into a polarization tester. Take the through capture, then rotate the emitting screen a quarter turn in its own plane and capture again. If the sample doesn't care about polarization, the two swings match. If it does — polarized sunglasses, polarizing window film, stressed plastic or laminate — the crossed capture passes far less light than the parallel one, and the drop measures how strongly the sample polarizes.

Notice what's missing: a direct capture. The reference for the crossed shot is the same sample at the other angle — the sample is its own control, so the camera chain, the geometry and even the pane's plain absorption all cancel. Two captures, one rotation, done.

Why polarization got its own card: physically, the polarization check is the transparency activity's optional third capture — same tool, same rig, same maths. But someone testing sunglasses would never think to open an activity called "glass transparency", and once there they'd be asked for a direct capture they don't need. So the check exists as its own activity, with just the parallel and crossed steps. Under the hood it's the same measurement with the same names in your catalog, whichever card produced it — a reading doesn't change identity based on which door you came in through.

The crossed capture also earns its place inside the transparency flow for a defensive reason: a pane that polarizes would quietly corrupt a plain transmission reading — the polarized emitter would be attenuated by an effect the single ratio can't see. Offering the quarter-turn check alongside is how the activity catches the confound instead of absorbing it.

What you'd actually use this for

Refusing to answer is a feature

The thread running through this activity is what it declines to do. No ratio without both captures. No ratio when the blink rate wasn't confirmed. No ratio when the reference swing is too small to divide by. No per-channel number where a channel wasn't driven hard enough. No silent re-aiming between captures. Each refusal is surfaced as a specific, fixable capture problem — because a measurement tool that guesses when its conditions aren't met isn't a measurement tool, it's a mood ring with a confidence interval.

The same-receiver ratio is one of several ways phone sensors become instruments once you build on relative features instead of absolute levels — the tools overview maps the rest.