I mostly agree with your assessment of what we should/shouldn't call open source for models but there is enough grey area to make the other side a valid position and not worthy of being dismissed so easily. I think there is a fine line between model weights and, say, bytecode for an interpreter and I think if you released bytecode dumps under any license it would be called out.
I also believe the four freedoms are violated to some extent (at least in spirit) by just releasing the weights and for some that might be enough to call something not open source. Your "freedom to study how the program works, and change it to make it do what you wish" is somewhat infringed by not having the training data. Additionally, gpt-oss added a (admittedly very minimal) usage policy that somewhat infringes on the first freedom, i.e. "the freedom to run the program as you wish, for any purpose".
You are free to look at every single weight and study how it affects the result. You can see how the model is architected. And you don't need training data to be provided to be able to modify the weights. Software can still be open source even if it isn't friendly to beginners.
I think you could say something remarkably similar about just releasing bytecode as well and I think most people would call foul at that. I don't think it's so cut and dry.
This isn't entirely about being a beginner or not either. Full fine-tuning without forgetting does really want the training data (or something that is a good replacement). You can do things like LoRa but, depending on your use case, it might not work.
I also believe the four freedoms are violated to some extent (at least in spirit) by just releasing the weights and for some that might be enough to call something not open source. Your "freedom to study how the program works, and change it to make it do what you wish" is somewhat infringed by not having the training data. Additionally, gpt-oss added a (admittedly very minimal) usage policy that somewhat infringes on the first freedom, i.e. "the freedom to run the program as you wish, for any purpose".