Here's an analogy. In some churches, the faithful will kneel in order to pray. Now I come along and say that praying 200 times per day will lead to better health. We put this to a test, and find it's true. But there are conflating factors - is it the praying, or the physical exercise through kneeling, or doing it in a church, or all of the above, which lead to the outcome?

A Kneelologist could stop, be happy that it works, and promote Kneelism as a healthy activity. But a non-kneelologist could point out that it's similar to calisthenics, which was already known to give similar positive results, is simpler to understand because it doesn't require the prayer component, and can done by people who are against prayer or don't have ready access to a church.

(Or for a real world example, the asanas from hatha yoga are used as exercise, and called 'yoga' even though yoga is a much broader topic.)

The scientific approach would address some important factors: 1) is the effect real and reproducible?, 2) when should be be used instead of other forms of treatment, and 3) what are the possible conflating factors and can we disentangle them to improve 1) and 2)?

Applying that to NLP, and making this up because I don't know the details. What if NLP is an incorrect synthesis of real-world observations that were already known at the time NLP was developed? In that, the ability of NLP to predict similar effects is not surprising. Other psychology models developed since Bandura's work in the 1960s also need to "predict" that behavior rehearsal can be an effective treatment.

Instead, what new predictions does NLP make which are different from other behavioral models? Can those predictions be tested? Or failing the ability to make new predictions, is it a simper model which it at least equally effective as other models in describing behavior?

That's where the science comes in.

NLP might work. But so might cognitive psychology, and with seemingly fewer worries about charlatans.

Yes you are right in the scientific approach.

What's more, NLP was not really developed. The techniques have been copied - or "stolen" - from other approaches, starting with Gestalt therapy. Therefore it's not a simple coincidence that you encounter similar techniques elsewhere.

Also, does NLP really need to bring novel ideas for it to be recognized? It does not promise anything like that, because that is not it's goal. NLP is about stealing what works to extend its "toolbox". That is also one of the points why it is considered being a pseudo science, "copying what works" is not scientific. In my opinion, not focusing on science is a major flaw of NLP. Therefore I hope that the NLP research organisation or others will change that.

Also, like you said, it is also possible that there are flaws in the copies, i.e. by making wrong conclusion from the observations. I would find it interesting to compare the NLP versions with the scientific versions to find the differences and maybe correct one or the other model after running some tests.

Btw. Do you know if depression or PTSD therapy based on the "scientific" psychoanalysis does yield results today?

"does NLP really need to bring novel ideas for it to be recognized?"

Of course not. It could be synonymous with "best practices." But even if it were identical to best practices, it uses set of unique terminology which makes it harder to understand. Why does it do that?

I do not know the answer to your last question.

That I do not know. However, that is not always the case. AFAIK Hypnosis after Milton H. Erickson is also part of NLP and similar terminology (if not the same) is used there.