I haven't changed my algorithm in 6 years, SVM with the kernel trick for nonlinear input spaces and supervising it is really effective though for this type of data. There might be better ones, but I think it would be a negligible difference. The key part is what are your valuing and feeding it IMO.
Harness isn't fixed AFAIK, but the conditions of the tracks are really bad. There's only one track (Meadowlands) that doesn't have bias. Harrah's track has been pretty broken since Sept of last year where you could blindly bet the 5 and you would have won ~$700 on a $2 bet.
I don't think the market caught up yet, since I just hit a 20/1 there. At that track like 83% of my bets are on the 5 since it opened a couple weeks ago.
Also hit a nice 8/1 on R1 for $715 and the $1380 in the 5th. It's because they don't re-invest the harness money back into the track so they fall apart.
The other issue is a lot of the horses are just horrible. There's so many racinos that need horses and they all get run into the ground. Some of them race multiple times a week. The quality would be better if the cards were smaller and less tracks, but until the rules change about racino's that will never happen.
A third issue is pool size. At Saratoga TBred you can bet 1k and barely move the pool because the WPS pools are ~500k. Most Harness are in 5-10k and 10k for exacta except for Woodbine (30k ish) and Meadowlands (50k-100k)
Some tracks like Yonkers I feel are unbeatable in WPS, but there's value in the exotic pools.