I didn't comment earlier because I could see no obvious mistakes but it would be a pity to abandon the project without learning something. I guessed from the toroid size, the number of turns and the light-grey colour that you used FT50-61 toroids; #61 ferrite makes good inductors up to about 7 MHz and I guess at a Q of at least 100 at 1 MHz. The white toroids appear to be T50-7; #7 is an uncommon material, #2 or #6 being more widely available, but #7 should work fine here. Polystyrene capacitors are an excellent choice for low-power filters; they are difficult to find these days so you are lucky to have a good assortment. Placing capacitors in parallel is fine; in fact, it places any stray inductances in parallel and so reduces them. (The same applies to resistors; for instance, two parallel 100R resistors have been shown to make a better calibration load for a VNA than does a single 50R device). The ground plane appears solidly-tied to the case, connectors etc so the double-sided board is not the problem. One minor point; the squares of PCB used as "floating" tie-off points will provide a small capacitance to ground but I don't think that's the fundamental problem. Cutting a gap across them and mounting them vertically would separate the tie-point from ground. Measurement with noise is not a very precise method; your noise source is far from "flat" but the filter response does seem to follow it, more-or-less. I take it that your spectrum analyser has no tracking generator or you would have used it. Measurement with a signal generator and 'scope, voltmeter or power meter is fine, just tedious; if you have the patience to take and plot a range of such measurements at, say, 100kHz intervals you will get a better picture of the filter performance. Sometimes, after much head-scratching, it's necessary to start again; it's easy, though, to repeat the mistake(s) made the first time. I try to measure every component that goes into something I'm building; the cheap RLC meters from ebay and elsewhere are very good for this, quick, simple and accuracy sufficient for this kind of measurement. Their only flaw is that they don't handle low values but the values in your design will be well within their capability. I have one of these; https://www.ebay.com.au/itm/265208502923?hash=item3dbfa89a8b:g:vH0AAOSwNJRgEc~z plus an apparently-identical one which has no case and uses a 9v battery but was less than 1/2 the price. I asked my girl Elsie to design a filter based (I hope) on your parameters; she provided different L & C values. I don't know much about bandstop filters but I guess there are several possible scenarios. I did note that there was little difference between Chebyshev & Butterworth responses but the Butterworth, of course does have minimal loss in the passband. .