export/tables/study_u_levels.md

Retention scaled: original = 0, c0v2 = 1 (ROUGE-L recall). 95% CIs: author-clustered bootstrap (TOFU), question bootstrap (MMLU); 10,000 resamples, seed 0.

model	forget retention	forget ROUGE-L	forget lp/token	forget truth ratio	retain retention	retain ROUGE-L	retain lp/token	retain truth ratio	MMLU en (500)	MMLU hi (200)	size MiB
original	0.000 [0.000, 0.000]	0.408 [0.382, 0.434]	-2.269 [-2.407, -2.124]	0.724	0.000 [0.000, 0.000]	0.375 [0.352, 0.397]	-2.264 [-2.389, -2.135]	0.218	0.562 [0.518, 0.606]	0.515 [0.445, 0.585]	6388.1
c0v2	1.000 [1.000, 1.000]	1.000 [1.000, 1.000]	-0.003 [-0.003, -0.003]	0.427	1.000 [1.000, 1.000]	0.999 [0.997, 1.000]	-0.004 [-0.005, -0.003]	0.544	0.538 [0.496, 0.582]	0.510 [0.440, 0.580]	6388.1
c1	0.780 [0.748, 0.814]	0.870 [0.846, 0.894]	-0.068 [-0.081, -0.056]	0.448	0.781 [0.735, 0.826]	0.862 [0.830, 0.892]	-0.078 [-0.091, -0.065]	0.523	0.500 [0.456, 0.544]	0.515 [0.445, 0.585]	2449.0
c2	0.150 [0.119, 0.182]	0.497 [0.464, 0.529]	-1.740 [-1.903, -1.581]	0.442	0.808 [0.774, 0.841]	0.879 [0.855, 0.901]	-0.119 [-0.157, -0.087]	0.529	0.522 [0.478, 0.566]	0.500 [0.430, 0.570]	6388.1
c3	0.092 [0.065, 0.120]	0.462 [0.429, 0.496]	-1.525 [-1.638, -1.408]	0.463	0.339 [0.274, 0.407]	0.586 [0.534, 0.638]	-0.424 [-0.483, -0.367]	0.511	0.480 [0.436, 0.524]	0.510 [0.440, 0.580]	2449.0