I'm trying to build a Butterworth LC bandpass filter to extract a 28.8 MHz signal for experiments, so I synthesized a 25-30 MHz Butterworth band-pass filter in Qucs and tweaked the component values to match the parts I have. It works fine in simulation. Unfortunately, when I built the circuit, it resonates strongly at ~28.8 MHz, and I can measure a 50 ohm impedance even if nothing is connected to the output. But I measured an insertion loss around 30-40 dB .

The simulated circuit, simulation result, actual circuit, PCB layout, and VNA measuments are attached. I omitted C107, L5, L6 - the last stage of the LC circuit (C4, L4 in simulation) and used some solder to fill the gaps. Dummy load R1 is not installed (not needed for two-port measurement), L7 is not installed, JP1 is closed. Capacitors are the old-fashioned ceramic disc capacitors, inductors are random 0603 inductors, probably multilayer construction.
I tweaked the capacitor and inductor values in the simulation to check the effects of extra parasitic elements, but an additional 20 pF or 20 nH cannot cause 40 dB of attenuation.
One category of possibility is cold solder-joints, badly etched board, or thin traces, but I've tested on a through-hole prototype board yesterday and seen similar attenuation, it would be extremely unfortunate to have the same cold solder joints in two different constructions. Another possibility is that the inductor has bad RF performance, but it's below 30 MHz and I don't think it matters at this frequency. It's also possible that the inductors are mismarked, but I hope it is not the case here...
Where should I start looking at?