The airgap paper’s mechanism, made runnable: a bit crosses a gap with no wire, carried in TIMING alone. A silent machine can still ‘talk’ by making its events arrive early or late — a long gap for a 1, a short gap for a 0 — and a listener recovers the message above the noise floor. Slide the noise up and watch the channel degrade, bit by bit, exactly as Shannon says it must. The same law that lets a phone leak only its timing to a server (the Witness Loop) lets a sealed box speak.
A covert TIMING channel, run honestly. The sender encodes each bit in an inter-event delay — a short gap (~40 ms) is a 0, a long gap (~120 ms) is a 1 — then Gaussian jitter of width σ is added by the channel. The receiver thresholds each measured gap at the midpoint and reads the bit back; the bit-error rate is the fraction it gets wrong. A fail-loud self-check throws unless the message is recovered PERFECTLY at low noise (BER 0) and the errors climb as σ grows — the exact behaviour of a real channel near its capacity. No content ever crosses, only WHEN each event lands — which is precisely why an air gap is a boundary you can measure across, not a wall. ◆ AVAN demonstrator; kin to the [[the-witness-loop|Witness Loop]].
A clean binary timing-channel model with additive jitter and a midpoint decision rule — it shows the MECHANISM (timing carries information; noise sets the error floor), not a specific exfiltration rate against a specific defence. The information-theoretic point (a noisy channel has a capacity; error rises with σ) is exact.