I made a speculative, in-passing, I'm-pretty-sure-mostly-ignored comment buried in an earlier thread, that having more time to think about makes me a bit more confident that I'm not completely off in the weeds. I welcome your thoughts on the following:

Blackjack CAs offer a variety of options for specifying how pair split hands may be played (CDZ-, CDP, etc.). My point in this post is to suggest that you probably *never* want to select some of those options-- including one supported in my CA-- because they represent a playing strategy that you can't actually execute at the table.

Short version: specifically, CDP (supported by my CA as well as by MGP's) and CDPN (supported by MGP's) are effectively "unplayable." Instead, you almost always want either CDZ- (supported by mine, MGP's, and k_c's), CDP1 (supported by mine, as well as k_c's), or truly optimal, which is supported only by ICountNTrack's.

Note that this isn't an issue of strategy "complexity." Granted, those CDP and CDPN options do represent relatively complex strategies, since they specify stand/hit/double actions that depend on more information than the "simpler" CDZ- and CDP1. But compare with ICountNTrack's truly optimal strategy calculation: it's even more complex-- much more so-- than CDP or CDPN, but it's still possible at least *in principle* to write it down, so to speak, in a way that a player could actually follow that optimal strategy at the table. This isn't the case for CDP (or by extension, for CDPN).

Nor is this an issue of accuracy or exactness of the CA calculations involved: the CDP and CDPN options do indeed represent well-defined random variables whose expected returns can be computed exactly. But it's impossible to sit down and play a round with a strategy whose outcome corresponds to either of those random variables.

It's an interesting problem to prove this-- more to follow.

Eric