Q factor is the line everyone skips in an RF cap datasheet
Had a bandpass filter, insertion loss was sitting about 2 dB off from simulation across the whole passband. Layout parasitics were accounted for, values were correct. Q factor was the issue. ESR at operating frequency was high enough to account for most of the discrepancy on its own.
Q degrades with frequency and most datasheets give you a single reference point that's usually nowhere near your actual operating conditions. NP0/C0G holds reasonably well as you go up in frequency, X7R doesn't, and in a resonant circuit that difference shows up directly in your measurements. Worth pulling the full ESR curve before selecting a part but not every manufacturer publishes it properly.