All modules can now be selected through "profile.txt" which looks like this. You can use comments as well. This introduces many branches per bit but lets me discover data depencies.

For example the match-model improves through sse which is senseless but shows that the estimate is improvable. Some context where slightly improved.

Code:

memhash=29
maxorder=8
counter_rate=32
mixer_rate=0.20,0.15,0.10,0.05,0.02,0.02,0.01
mixer_init=0.10
mixer_sse=0
mixer_sse_trust=0.60
mixer_sse_rate=0.010
counter_sse=0
counter_sse_trust=0.60
counter_sse_rate=0.010
match_model=1
match_win_size=22
match_hash_size=18
final_mix_rate=0.001
final_sse=1
final_sse_trust=0.75
final_sse_rate=0.020

intermediate sse gets more and more useless as the probability estimates gets better.

Hitting around 11.150.xxx on sfc and 20.912.xxx on enwik8.

order2 + match-model gets around 250k on book1