Being almost correct can be more dangerous than being obviously wrong.

An obvious error announces itself and gets corrected. A near-correct one passes every casual check, gets built upon, and quietly propagates into everything downstream. By the time the discrepancy is large enough to notice, it has been load-bearing for a long time, and unpicking it costs far more than the original mistake ever would have.

Be more suspicious of results that look plausible than of ones that look wrong. When something comes out close to what you expected, that is the moment to verify rather than relax. Ask what would be different if it were subtly off, and go check that specific thing.