The phase range midpoint reduced the first certified wait for a 7 Hz filter after simulated gaps in nine SEEG rest records with known states and voltage limits

We tested a phase reader after simulated loss of SEEG samples. We reused nine public rest records and 36 fixed bipolar pairs from OpenNeuro ds006065 and ds005670. The target was a 7 Hz causal complex filter state. We computed the exact phase range allowed by independently bounded missing real samples. Its midpoint gives the smallest possible maximum phase error. Under the same voltage limits, it admits every phase certificate that the zero-filled state admits. The main test had 4,140 isolated requested 25 ms gaps, a 15 degree limit, exact known pre-gap states, and a voltage cap twice the early maximum absolute centered input. The mean first certified grid wait fell from 292.72 to 282.39 ms. The mean reduction was 10.33 ms, with a descriptive record bootstrap interval of 9.60 to 11.32 ms. All nine record means improved. A denser grid gave a 10.11 ms reduction. At 100 ms, acceptance rose from 0.507% to 0.870%. All 165,600 accepted readouts whose missing samples met the cap had error below 15 degrees. Two gaps exceeded the cap and gave eight accepted errors above the limit. On jointly certified readouts, the equal pair and record mean recorded error rose from 0.379 to 0.643 degrees. A smaller worst-case bound does not ensure a smaller error on a particular signal. The proof concerns this filter phase under the stated voltage and state assumptions. No neural phase truth, decoding score, or stimulation outcome was tested. Fresh public retrieval matched all nine voltage extracts exactly. A second full analysis matched all numerical outputs. The package has the proof, online reader, fixed plan, data, source checks, results, three figures, and reproduction steps.

phase estimation · SEEG · signal processing · missing samples · uncertainty bounds