insertion-order confound: both gold paragraphs in the same 50-doc block for 98.3% of questions (chance ~4.2%)

=== insertion-order: 24 blocks ===
router                  L   p        beam=2        beam=4        beam=6        beam=8       beam=12       beam=16
centroid                1   -    60.0/ 85.0    66.7/ 91.7    68.3/ 95.0    70.0/ 97.5    70.0/100.0    69.2/100.0
centroid+radius         1   -    60.0/ 85.0    65.8/ 90.8    66.7/ 92.5    70.0/ 96.7    71.7/100.0    69.2/100.0
landmark_max            8   -    71.7/ 98.3    70.8/100.0    70.0/100.0    69.2/100.0    68.3/100.0    68.3/100.0
landmark_pmean          4  13    70.0/ 95.8    71.7/100.0    70.0/100.0    70.0/100.0    69.2/100.0    69.2/100.0
landmark_pmean          8   5    65.8/ 92.5    70.0/ 96.7    70.0/ 97.5    70.8/ 98.3    70.0/ 99.2    69.2/ 99.2
landmark_pmean          8  13    70.8/ 98.3    71.7/100.0    70.8/100.0    70.0/100.0    68.3/100.0    68.3/100.0
landmark_pmean          8  21    71.7/ 99.2    71.7/100.0    70.8/100.0    70.0/100.0    68.3/100.0    68.3/100.0

=== random-order: 24 blocks ===
router                  L   p        beam=2        beam=4        beam=6        beam=8       beam=12       beam=16
centroid                1   -     5.0/  5.0     9.2/  9.2    16.7/ 18.3    18.3/ 20.8    21.7/ 30.8    21.7/ 34.2
centroid+radius         1   -     3.3/  3.3     5.0/  5.0    13.3/ 15.0    17.5/ 20.0    20.8/ 28.3    20.8/ 33.3
landmark_max            8   -     6.7/  6.7    12.5/ 14.2    17.5/ 24.2    18.3/ 28.3    20.8/ 35.0    20.8/ 35.8
landmark_pmean          4  13     6.7/  6.7    13.3/ 15.0    15.8/ 19.2    20.0/ 26.7    20.8/ 32.5    20.8/ 35.0
landmark_pmean          8   5     5.0/  5.0    12.5/ 13.3    15.8/ 18.3    19.2/ 23.3    20.0/ 29.2    20.8/ 32.5
landmark_pmean          8  13    10.8/ 10.8    16.7/ 17.5    18.3/ 22.5    21.7/ 29.2    20.8/ 35.0    21.7/ 37.5
landmark_pmean          8  21     8.3/  8.3    15.8/ 16.7    19.2/ 24.2    20.8/ 28.3    20.8/ 34.2    20.8/ 35.8

=== cluster-aligned k=48: 42 blocks ===
router                  L   p        beam=2        beam=4        beam=6        beam=8       beam=12       beam=16
centroid                1   -    55.8/ 75.0    65.0/ 89.2    66.7/ 93.3    66.7/ 95.0    67.5/ 98.3    68.3/ 99.2
centroid+radius         1   -    56.7/ 75.8    65.0/ 89.2    65.8/ 90.8    66.7/ 95.0    67.5/ 98.3    68.3/ 99.2
landmark_max            8   -    61.7/ 81.7    67.5/ 95.0    67.5/ 97.5    68.3/ 99.2    68.3/ 99.2    68.3/ 99.2
landmark_pmean          4  13    58.3/ 79.2    64.2/ 89.2    65.0/ 92.5    68.3/ 96.7    68.3/ 99.2    68.3/ 99.2
landmark_pmean          8   5    60.0/ 78.3    66.7/ 90.8    67.5/ 95.8    68.3/ 97.5    68.3/ 99.2    68.3/ 99.2
landmark_pmean          8  13    65.0/ 85.8    65.8/ 93.3    67.5/ 95.8    68.3/ 99.2    68.3/ 99.2    68.3/ 99.2
landmark_pmean          8  21    64.2/ 85.0    65.8/ 93.3    67.5/ 96.7    68.3/ 99.2    68.3/ 99.2    68.3/ 99.2

cells: evidence-recall@4 / gold-block-coverage (%). exhaustive = 68.3
