HOW WE TEST
The decoders in Radio Silence make claims that are easy to say and hard to earn. Live text off the air. Nothing invented when the band is empty. This page explains how we hold ourselves to them.
REAL AIR, NOT LAB TONES
Synthetic test audio is clean, obedient, and nothing like shortwave. So our reference library is recorded off the air, through the same public receivers the app tunes. Marine weather bulletins from DWD Pinneberg give us always-on RTTY at 50 baud. The NCDXF beacon network gives us CW callsigns that step down in power every ten seconds, a built-in ladder from strong to barely there. Every clip carries real fading, real static, and real interference, because it was radio before it was a test.
EVERY CHANGE ANSWERS TO THE LIBRARY
Decoder work here follows one rule. Measure first, change second, measure again. Each candidate change is replayed against the whole recording library and scored on what it actually recovered, checked against independent open source decoders run on the same clips. If a change helps one recording and quietly hurts three others, the library says so and the change does not ship. Several promising ideas have died exactly this way, and the numbers that killed them are kept on file.
SILENCE MUST STAY SILENT
The library includes control recordings of empty frequencies. Real band noise, real static crashes, no signal anywhere. The decoders are required to print exactly nothing on every one of them. A decoder that invents text out of noise is worse than no decoder at all, so this is the one test where the passing score is zero.
THE SECOND LISTENER
The hardest recordings are the ones where the signal spends most of its time near the noise. For those the app runs a second decode layer that buffers whole seconds of audio and studies them as one piece, finding the timing grid a character-by-character decoder loses. On the weakest recordings in the library it more than doubles the text recovered. And when even that layer is not confident, it prints nothing. See the rule above.
WHAT WE DO NOT PROMISE
Shortwave is physics before it is software. A path that has faded below the noise floor defeats every decoder ever written, including ours. What we promise is honest behavior at the edge. Print what the air actually carried, mark uncertainty by staying quiet, and never dress noise up as copy.