We audit 14 graduate-level statistics courses across 11 universities and find that statistical power receives an average of 23 minutes of instruction, compared to 4 hours for p-values and 6 hours for ‘assuming normality without checking.’
We introduce BenchmarkMax-9000, a rigorous evaluation suite on which our model achieves state-of-the-art performance, outperforming all baselines we selected.
We formalize the practice of adding data points until p < 0.05 as ‘adaptive sequential analysis’ and demonstrate that it achieves the desired result in 94.7% of cases given sufficient patience.
We track 312 sprint retrospective action items across 9 organizations and find that 89% are never completed, while 100% are re-generated in the following retrospective.
We identify 43 self-sustaining citation rings in machine learning literature in which a closed group of authors cite exclusively one another, achieving collective h-indices that exceed their individual contributions by a factor of 6.4.
We analyze 847 consecutive standup meetings across 14 engineering teams and find that 97.3% of information exchanged was available in Slack prior to the meeting.