Rev B (majority-vote receiver)

Rev B decides each received data, parity, and stop bit by a 2-of-3 majority vote across three oversample points straddling the bit center, so a noise glitch narrower than one oversample period cannot flip a bit or fake a framing error. The change is verified at RTL by sample-point glitch tests that fail on rev A and pass here, by a measured clock-mismatch envelope of -4.63% to +5.56% against the documented +/-4% contract, and by the full pin-level suite.

Rev B is not fabricated. Its extra cells push the 1x2 tile past its routing margin at 77% utilization, so the proven rev A netlist remains on the shuttle.