The replication crisis is a crisis in that it wastes a lot of time and money: a cool but wrong result is published, and a bunch of labs rush to build on the work. Most fail, shrug, and move on with something else. A few dig hard and either discover it's really genuinely wrong (and may or may not publish that), or that there's some subtlety to the phenomenon that wasn't originally appreciated and some new findings come out of it.

It happens less in physics and chemistry, because the systems are simpler, and the expectations are clearer, but anything from molecular biology on up in complexity is affected.

But it's not a crisis in that experiment-heavy sciences accumulate oodles of wrong ideas that get increasingly entrenched as dogma and harder and harder to unseat. Even at current levels of replication-crisis, the overall scientific approach is way way way within the safe zone where errors are detected and fixed. It's just slower and more expensive than it should be, because too much gets published with too little checking, and too little gets published that reports that the first people stuffed something up.

Drug companies hate it, because it means they can't just read a cool paper for free (or maybe $50, if it's not open access) and then start running towards a billion-dollar product. Checking things out doesn't cost them much compared to full blown drug development, but it does slow them down.

Grad students hate it, because their PI just got really excited about this result out of so-and-so lab and everyone knows there's so much hype out of that lab but somehow they always get published in the big journals and it's going to be so much work and it's like 50-50 whether it's going to pan out (does anyone know the lead author?) and ha-ha, no, there's no way I can work on this as my "backup project"; it's too much.

But from a philosophical perspective it's totally cool.

Some things stay wrong or undetermined for a long time (hey there, amyloid fibrils!), but that usually happens when it's not experimentally tractable to really address the question thoroughly. Um. Maybe that's not totally true with amyloid fibrils. But people get to it eventually.

And so we add cumulatively to the store of reliable human knowledge, with somewhat lower-fidelity labeling of when something is reliable than we'd like, and not as fast as we'd like, but definitely very much still doing it.