Finite sample tests support period noise beyond a cubic trend in the 2014-2026 ASASSN-14ko peak times

We test if a smooth change of period decay removes the evidence for random period changes in ASASSN-14ko. We use 30 prior peak times and their reported errors, with two peak definitions. An exact restricted Gaussian likelihood allows independent white timing scatter and integrated random period kicks. We calibrate two nested noise tests with 999 null simulations each, for each peak definition. Period noise remains supported after a cubic timing trend is removed: simulated tail probabilities are 0.005 and 0.001, with limited Monte Carlo precision. The fitted kick scales are 0.72-0.97 days per orbit with the cubic trend. A training-only test also gives the period-kick model higher joint forecast scores than a cubic model for the nine later peaks. However, full profile maximum likelihood places the kick scale at zero. Two broad modes carry most of the covariance sensitivity. The evidence thus depends on the treatment of the unknown mean and noise. These tests concern the flare clock under Gaussian assumptions. They do not identify orbital kicks or exclude changes of emission delay. No new photometry is used.