Lowering the TF level to 33 bit and setting tests saved to 2 made it do something. So that means with realistic values it's better to do no P-1?

I guess I'll just continue sieving and LLRing from the low end then.

Thanks everybody for your help!

3M starts needing P-1 at 42 bits, 46 bits with 2 tests saved
4M starts needing P-1 at 44 bits, 48 bits with 2 tests saved
5M starts needing P-1 at 45 bits, 49 bits with 2 tests saved
Looks like my tests were with 40 originally and I overestimated how fast the start bits would go up/underestimated the sieve depth. 2 tests seems to put the start bits up 4.
Sorry for the time wasted on this. At least we learned something.

Sieving is closing in on 800T. I have started LLRing from the small end of the candidate file. So I am officially reserving R6 to n=2.5M for a start.

R6 tested to n=2.5M, no prime. Continuing to n=3M

R6 tested n=2.5M to 2.8M, no prime. Continuing...

R6 tested n=2.8M to 3M, no prime. Continuing...