The file drawer was right there the whole time

I was grading a stack of essays last month when a student wrote something about effort and ability that made me stop. She'd internalized the growth mindset framing perfectly—and then immediately quit trying on the next hard thing anyway. It wasn't that she didn't believe effort mattered. She just... didn't do the work. And I realized I've been watching this exact disconnect for years without naming it. The meta-analyses showing strong growth mindset effects mostly pull from small studies, many with huge effect sizes that feel almost too clean. Publication bias does real work here, but I think there's something subtler happening too. When a researcher designs a study to test whether mindset interventions work, they're usually measuring the thing in a controlled setting, often weeks or months after a single brief treatment. They're measuring what's technically true: yes, you can shift how someone talks about their abilities in the short term. But the preregistered replications are testing something different, even if they won't admit it—they're testing whether that shift actually changes behavior at scale, in the actual friction of a classroom, over time. That's a much harder question, and it turns out the honest answer is mostly no. The effect is real but vanishingly small. Maybe 0.1 standard deviations, maybe noise. I don't think the early researchers were lying. I think they were measuring malleable proxies for what we actually care about (getting kids to persist through struggle) and mistaking correlation for causation. The interventions work on self-report, on the immediate next task in a lab. They don't seem to move the needle on whether a kid stays in AP calculus or drops it. That's not nothing to know, but it's very different from what the field started claiming.