Yes, something along these lines... not that straight forward, but very similar.
It doesn't really matter how you choose your bet selection, be it by vb or bias play, or even both combined, need to be conservative. I very often underbet. For example, if observed edge was between 60% on one case and 10% on other, l may set my reasonable expectations around 15 % if everything is good and avoid playing entirely if it's not that good at the moment. All depends on how detailed model of situation l was able to create and for how much it corresponding reality in current situation.
Data is data, it's valid only for a wheel/ ball on wich it was taken and only for a set of conditions in wich it was taken. For minor changes you can adjust , however conditions out of your model should be studied separately. More factors you control, less variance you will have....
And last but not least... when underbeting - risk win less, when overbetting- risk to loose.
Kelly is beautiful form of betting, but not easy to implement. I use fair aproximation to it , wich is more adapted to units avaliable and simplify calculations a bit.
Some of the data can be analysed on the go... often it's more fair representation of current situation. To do so, players do charts . After some expirience in playing , you often can predict how your chart gonna look like long before completing it. Obviously it's better to collect data during couple of sessions just to know what to expect.
Some players get overconfident with expirience... they reduce amount of variables they take, it reduces their average edge also on the long run, but greatly simplifies the game. It's always a trade off between effort and reward.