The solution to poor results from scientific work is better scientific work, not throwing out the scientific model. The issue described is akin to testing some

The solution to poor results from scientific work is better scientific work, not throwing out the scientific model. The issue described is akin to testing some new code with unit, integration, and end-to-end tests, and running it for some time in a dev environment, but not running it in a QA environment that mirrors production, and not monitoring it for issues when it does get deployed to production. The response there wouldn't be "the test/devops model isn't the correct one to use in the first place", it's that the testing and QA was inadequate because it only looked at an idealized model of reality and didn't consider "where the rubber meets the road" (literally). Testing something in isolation is insufficient, since you need to test the downstream effects and potential complex interactions with everything else in the environment. There is no testism problem; only a poor testing problem. Unless, as you say in your last sentence, one is just arguing about definition semantics - e.g. if "QA" falls under "testing", or, here, if "real-world observation" falls under "scientific testing". I'd argue both do. I think your concerns might be better phrased as "lab model" and "lab-ism".