I've been doing some experimentation as well and met the same issue. The way I've been dealing with it is to use a CW tag before and after the mode (as well as sometimes alongside my data mode). In it I put my callsign, and a tiny url (as its morse)
e.g. if my experiment data was at google.com, i nip to tinyurl and get a tinyurl:
http://tinyurl.com/2tx, so I would transmit:
[CALLSIGN] EXPERIMENT SEE - HTTP TINYURL COM 2TX [CALLSIGN]
and hopefully people understand that without the punctiaton. The page itself is generally a blog page containing information on what I'm doing - not necessarily anywhere you can download a file to decode or instructions, but importantly I put a contact form or a way of getting in touch if I'm causing interference.
I currently pre-code the audio files that I transmit, and can overlay that morse alongside the data blob so it goes out like a margin notes.
The rules (in the UK) say you can't transmit secret codes, and the essence of my work not being public isn't that I'm being secret, just that I'm not ready to publish the source (thats arguably an interpretation). In practice it wouldn't take long for someone to analyse and decode the message as mine are all PSK based of underlying data.
Software testing with generated noise etc is good, but doesn't always let you test reality. I probably do 99% of my tests with just audio connections between overplayed with static etc, which lets me tinker without having to make real radio connections between people.