1. Incomplete gamma self-tests PASS P(1, 1) = 1 - e^-1: got 0.63212, want 0.6321205588285577 (tol 1e-12) PASS P(0.5, 2) = erf(sqrt 2): got 0.95450, want 0.9544997361036416 (tol 1e-10) PASS P(3, 2.5): got 0.45619, want 0.4561868841166703 (tol 1e-10) PASS P(0.2, 50) ~ 1: got 1.0000, want 1 (tol 1e-12) 2. Printed parameters reproduce the papers' printed class tables PASS EW2006 share Nes 0-1: got 0.19464, want 0.19 (tol 0.015) PASS EW2006 share Nes 1-10: got 0.13560, want 0.14 (tol 0.015) PASS EW2006 share Nes 10-100: got 0.22556, want 0.23 (tol 0.015) PASS EW2006 share Nes 100-1000: got 0.31388, want 0.31 (tol 0.015) PASS EW2006 share Nes 1000-10000: got 0.13005, want 0.13 (tol 0.015) EW2006 share above Nes 10,000 (not printed by the paper): 0.0% PASS Boyko2008 share |s| 0-0.0001: got 0.27889, want 0.279 (tol 0.01) PASS Boyko2008 share |s| 0.0001-0.001: got 0.14676, want 0.147 (tol 0.01) PASS Boyko2008 share |s| 0.001-0.01: got 0.21897, want 0.219 (tol 0.01) PASS Boyko2008 share |s| > 1e-2: got 0.35538, want 0.355 (tol 0.01) PASS Boyko2008 mean s = a*b/(2N), printed -0.029 (p. 4): got 0.029427, want 0.029 (tol 0.0005) Kim2017 N_A calibrated on Table 3 bin 1: 12300 diploids (a calibration, not a printed or independently recovered number) PASS Kim2017 share |s| 0-0.00001: got 0.23700, want 0.237 (tol 0.01) PASS Kim2017 share |s| 0.00001-0.0001: got 0.12656, want 0.127 (tol 0.01) PASS Kim2017 share |s| 0.0001-0.001: got 0.19217, want 0.192 (tol 0.01) PASS Kim2017 share |s| 0.001-0.01: got 0.26579, want 0.266 (tol 0.01) PASS Kim2017 share |s| >= 1e-2: got 0.17848, want 0.178 (tol 0.01) 3. Share of new amino-acid mutations below a drift threshold, by assumption (each paper's own s convention, so the rows are not on one biological scale; threshold 1/(4Ne) and 1/(2Ne); Ne 10,000 and 20,000) (the threshold is an approximate, graded drift scale, not a line below which selection is impossible) EW2006 (s het, Ne 1e4 built in) 1/(4Ne=10k) 2.5e-5: 14.2% 1/(2Ne=10k) 5.0e-5: 16.6% 1/(4Ne=20k) 1.3e-5: 12.1% 1/(2Ne=20k) 2.5e-5: 14.2% Boyko2008 AA (s het) 1/(4Ne=10k) 2.5e-5: 21.6% 1/(2Ne=10k) 5.0e-5: 24.6% 1/(4Ne=20k) 1.3e-5: 19.0% 1/(2Ne=20k) 2.5e-5: 21.6% Kim2017 1000G (additive s, N_A calibrated) 1/(4Ne=10k) 2.5e-5: 28.1% 1/(2Ne=10k) 5.0e-5: 32.0% 1/(4Ne=20k) 1.3e-5: 24.7% 1/(2Ne=20k) 2.5e-5: 28.1% Range over fits and assumptions: 12.1% to 32.0% 4. Beneficial shares, kept apart by how the beneficial class was decided 4a. Random samples, every mutant classified by a significance test VSV, 48 random substitutions (Sanjuan 2004 Table 1): beneficial 4.2% of all, 5.7% of non-neutral; harmful+lethal:beneficial 16.5:1; Sanford's 1e6:1 is 6.1e+4 times more extreme f1 phage, 100 substitutions, BH-corrected (Peris 2010 p. 606): beneficial 2.0% of all, 4.3% of non-neutral; harmful+lethal:beneficial 22.5:1; Sanford's 1e6:1 is 4.4e+4 times more extreme 4b. Random samples with no significant beneficial call (bound on a detectable call, iid assumption) TEV, 66 substitutions, all mutants (Carrasco 2007): 0 of 66; 95% upper bound on P(significant beneficial call) = 4.4% (rule of three 4.5%) E. coli, 226 random insertions, all mutants (Elena 1998 p. 354): 0 of 226; 95% upper bound on P(significant beneficial call) = 1.3% (rule of three 1.3%) PASS TEV zero-event bound, n = 66: got 0.044375, want 0.0444 (tol 0.0005) PASS Elena zero-event bound, n = 226: got 0.013168, want 0.0132 (tol 0.0005) 4c. Beneficial classes not decided by a corrected test (provisional; no ratio priced) Yeast tRNA (Li 2016): about 2.1 of 207 called beneficial at nominal P < 0.05; if the ~118 non-significant mutants were truly neutral, about 2.9 beneficial calls would be expected by chance alone. That makes the 1% provisional; it does not show which calls are false. Nominal ratio 42:1, not used as a counted ratio. Influenza (Visher 2016): growth bands sum to 89 of 90 viable; the 90th viable mutant (HA-8) could not be measured (p. 7). 7 of 89 measured viable mutants ( 7.9%) fall in the 1.05-1.15 band, which the authors did not test for significance. No ratio priced. PASS Visher bands sum to 89 of 90 viable: got 89.000, want 89 (tol 0) 4d. Fitted or line-derived beneficial shares (model-conditional) Yeast diploid MA lines (Joseph & Hall 2004), fitted 5.75% of fitness-altering mutations: implied ratio 16.4:1 (a model parameter, not a count of classified mutations) E. coli insertions, LTEE ancestor (Couce 2024 p. 9): deleterious:beneficial 3.0:1 among insertions (not point mutations) Burkholderia M9MM (Dillon 2016 p. 1231): U_D/U_B = 3.2:1, i.e. 13 worse lines to 4 better lines; derived under the paper's one-mutation-per-line assumption 4e. Wloch 2001 excluded: ascertained harmful class. Lethal share WITHIN the 69 ascertained spontaneous mutations: 29.0% 5. Beneficial rate divided by an ASSUMED total genomic rate of 3e-3 (Sniegowski & Gerrish 2010, for E. coli and yeast) Gerrish & Lenski 1998, E. coli, model fitted to three sweeps (their own division, p. 132): U_b 2.0e-9 -> 6.7e-7 of new mutations (1 in 1,500,000) Imhof & Schlotterer 2001, E. coli, sweeps with s >= 0.01: U_b 4.0e-9 -> 1.3e-6 of new mutations (1 in 750,000) Perfeito 2007, E. coli, as restated by Sniegowski & Gerrish 2010: U_b 2.0e-5 -> 6.7e-3 of new mutations (1 in 150) Desai, Fisher & Murray 2007, yeast, model fit: U_b 2.4e-5 -> 8.0e-3 of new mutations (1 in 125) Levy 2015, yeast, s > 5%: U_b 1.0e-6 -> 3.3e-4 of new mutations (1 in 3,000) Spread between the Imhof and Perfeito figures: 5000-fold (different strains, adaptation histories, population sizes, models and detection limits) 6. Creationist working values against the same yardsticks Basener & Sanford 2018: working fraction 0.001 vs 1/(1e6+1) = 1.00e-6 implied by the cited ratio: about 1000 times (approximate; assumes no separate neutral class, as in their model) Brewer 2013 input: neutral 10.0%, deleterious 89.1%, beneficial 0.9% Sanjuan 2004 random set: neutral 27.1%, deleterious+lethal 68.8%, beneficial 4.2% ratios (Sanjuan/Brewer): neutral 2.7x, beneficial 4.7x All checks passed.