Table 3.

Non-parametric statistical tests comparing terminal performance at DAP and post-DAP for curious model variants.

Mann-Whitney U-Test (n = 128, α = 0.025, Bonferroni corrected)
Validation loss		DAP (T = 30, 000)	Post-DAP (T = 60, 000)
C/B < C/PE	Statistics	6558.0	5911.0
	p-value	0.0029	5.9E-5
C/B < PG/IRS	Statistics	6275.0	5062.0
	p-value	0.0006	6.4E-8

Following Table 2, even though the boredom score came close to other curious variants (C/PE and PG/IRS), the boredom variant still outperformed the other two on statistical grounds.