What White Rabbit Measures: Round Trips, Local Phase and a Calibrated Prior

White Rabbit, standardized as the High Accuracy profile of IEEE 1588-2019, is often described as measuring one-way link delay in hardware. It does not, and no in-band procedure can. White Rabbit measures three things: the round trip, to picoseconds; local quantities at each node, such as the phase between two clocks on one board and the deserializer bitslide; and nothing else across the link. It obtains the one-way delay by splitting the round trip with a structural model whose parameters, the fixed transmit and receive delays of each device and the relative delay coefficient alpha of the fiber, are supplied by calibration performed on co-located equipment. The calibration's absolute step rests on a stated convention that the reference device has no asymmetry. This is sound engineering, and it explains the sub-nanosecond results. It also fixes what the accuracy depends on. Every picosecond by which the true asymmetry departs from the calibrated one becomes half a picosecond of time error, and because a change of asymmetry that preserves the round trip leaves every exchanged timestamp unchanged, no in-band mechanism can detect it. We set out how White Rabbit works (Part I), restate the two-way bound on in-band offset estimation with a self-contained proof (Part II), and derive the operational consequences for re-patching, component replacement, holdover and silent failure (Part III), with five recommendations for specifications that adopt 1588 High Accuracy.