(0) full reference 32,561 rows, stattest=chisquare for every column, no shift, 2000 windows n=50 dataset 0.2% Educat=8% Marita=8% Native=23% Occupa=6% Race=6% Relati=5% Sex=6% Workcl=5% n=100 dataset 0.4% Educat=7% Marita=11% Native=19% Occupa=7% Race=5% Relati=5% Sex=4% Workcl=8% n=200 dataset 0.4% Educat=6% Marita=6% Native=14% Occupa=8% Race=6% Relati=5% Sex=5% Workcl=8% (0b) reference 1000 rows, no shift, window growing: Relationship (no rare category) n=50 alarm 7.8% n=200 alarm 10.7% n=500 alarm 20.8% n=1000 alarm 34.7% (1) reference 1000 rows, no shift, alarm rate at p<0.05 column n Evidently style expected from full data no unseen value trials expected<1 Race 50 8.8% 5.5% 8.8% 3000 40% Race 100 10.6% 5.3% 10.6% 3000 24% Race 200 12.1% 5.4% 12.1% 2999 2% Workclass 50 8.6% 5.0% 6.5% 2933 7% Workclass 100 11.7% 7.3% 8.1% 2881 7% Workclass 200 18.0% 9.0% 10.4% 2746 7% Education 50 15.1% 8.4% 13.4% 2941 43% Education 100 17.1% 7.1% 14.3% 2903 16% Education 200 25.7% 5.8% 20.8% 2812 8% Native_country 50 51.5% 22.1% 20.6% 1833 94% Native_country 100 70.8% 18.2% 21.8% 1122 90% Native_country 200 87.8% 13.1% 18.8% 451 80% (2) five categories, reference 1000 rows, no shift, Evidently style test on a large current current 1000: alarm 32.1% current 10000: alarm 92.8% current 100000: alarm 99.9% (3) five categories, 0.5pt moved between two categories, two samples of size n, chi2_contingency, 1000 runs population Jensen-Shannon distance 0.0085 n=10000 median p=2.1e-01 share p<0.05=21.8% mean distance=0.0123 n=100000 median p=1.9e-06 share p<0.05=99.5% mean distance=0.0090 n=1000000 median p=2.3e-62 share p<0.05=100.0% mean distance=0.0086 no shift, n=1000000: share p<0.05=4.1%