Regional use of the International Reference Ionosphere (IRI) requires knowing how much historical calibration reduces local errors and which conditions remain poorly represented. We examine these questions at Hainan Fuke using 2014–2020 observations and a fixed IRI-2026 configuration. The targets are the F2-layer critical frequency (foF2), which reflects peak electron density, and peak height (hmF2). Corrections learned from 2015–2018 are evaluated in later years. In 2020, a simple month-hour correction reduced foF2 mean absolute error from 1.137 to 0.997 MHz, slightly below the error of the random forest of 1.013 MHz. For hmF2, the forest reduced error from 28.634 to 21.754 km, compared with 22.941 km for the simple correction. The forest's additional benefit therefore differs between the peak parameters. Across recorded spread-F classes, which describe broadened ionogram echoes, estimated percentage gains in foF2 exceeded those without spread F. Yet the corrected error during strong range spreading remained 3.622 MHz, compared with 1.037 MHz without spreading. Height showed its largest estimated gain without spread F. A separate retrospective update using the preceding 15-minute residual gave errors of 0.256 MHz and 10.851 km. For this Hainan record, method choice depends on the target and on whether average absolute error or mean bias matters most. Strong spreading retains substantial frequency errors despite larger relative gains, making the remaining errors for each parameter and state important for regional use. Recent observations further improve local estimates, although an operational forecast horizon remains untested.