Observability as Openness: Finite-Stage Verification and How Non-Robust Structure Registers

Physics reports its exact null results as bounds. We derive from the apparatus the form of what a measurement can certify, of which that practice is the idealized expression. If a measurement stops in one of finitely many separated record states, reached continuously from the configuration, the content of a record, the set of admissible configurations that could have produced it, is open, and on a connected space a discriminating apparatus must sometimes return no verdict (Lamport’s principle); for independent noisy records, a theorem of Genin and Kelly gives the same openness for verifiable hypotheses. Two consequences follow, known in other forms from computable analysis and learning theory: what a registration certifies holds on an open set (O1), and a condition that fails at admissible configurations arbitrarily close to the true one is never certified (O2), while one that the admissible set makes hold on a neighbourhood, as for quantized values, is certified exactly. We then ask how much non-robust structure a registration of given precision delivers. At the golden Siegel parameter of z² + c, a Taylor-model enclosure and certified witnesses show that a bracket of radius ε on parameter and initial state certifies exactly the first K Fibonacci closest returns of an orbit in the disc, with K within one rung of ½ log_φ(1/ε) at the six widths from 10⁻⁴ to 10⁻⁹. The record structure rests on a ladder theorem for Siegel discs.