It's hyperbole to the level that obfuscates, unfortunately. 50% of psych findings being wrong doesn't mean "right all the time except in exotic edge cases" like pre-quantum physics, it means they have no value at all and can't be salvaged. And very often the cause turns out to be fraud, which is why there is such a high rate of refusing to share the raw data from experiments - even when they signed agreements saying they'd do so on demand.