Just do normal splits, do lots of cross solves, then lots of cross+F2L solves. Take the cross solves on their own, and for F2L, take the entire time you got, and take away the cross time to get how much time you spend on pairs. I don’t think you’re looking into OLL and PLL yet, so just time these separately.
I put in your multiphase times anyway, your F2L is good, but cross, OLL and PLL need a little work overall.
Generic advice:
Plan out cross in inspection, if you can’t yet, work on doing 2 pieces, then 3 pieces, etc. Cross should be done on bottom and rotationless if possible. Lookahead during cross into the first pair, if you’ve planned out the cross, you don’t need to look at what you’re doing and can focus on your F2L pair which you’ll start solving after cross. Know all 42 optimal algs for F2L, they should be fast and easy to recognise for you. Again lookahead whenever possible, you don’t need to look at your F2L pair if you know the alg, it’s in your muscle memory, so you can use that to start looking for the next pair. Slow down a little if you have to, try to turn consistently, bursts of high TPs and pauses are worse than normal, consistent TPS. You don’t need full OLL and PLL for sub 20, but I recommend PLL and OLL if you want. OLL and PLL should be below 3 and 4 seconds respectively for sub 20. Train specifically recognition and execution. Make sure you have good algorithms. Try not to rotate too much, use U moves to look at the top layer. Plus, consistent turning speed in F2L before TPS. Once you’ve got consistent turning (which means you have good lookahead), it’s easy to speed up turns, but not the other way. And PRACTISE. LEARNING THIS IS USELESS IF YOU CAN’T PRACTISE IT.
Split data for sub 20 is:
https://docs.google.com/spreadsheet...53dy_J2WLXMY9RTnEep2efBPWI/edit#gid=578524837