Remind us: How to interpret late-season CDG changes?

[JonE]

Can someone remind us how to understand what happens to the CDG projections and values at this point? I know that small stats can create huge swings in value (e.g. a pitcher projected for 3 saves in an only league can have an R$ of $27 and 2 saves means $19), but I'm not sure what to make of the fewer number of players projected. For instance, my NL-only league website tells me there are 18 free-agent catchers projected for positive AB, but the CDG says there are only 3. Should I really just stop using it altogether?

[Whistler]

Not a tech guy, but predicting the next 16 days stats would have wide error bars, even if the PT were perfect, which it's not.

[JonE]

Of course I understand about the error bars. But to those of us who are still in it, we want some guidance on whether (to stick with my example) Luis Torrens or Sean Murphy are going to get more PT than Freddy Fermin or Adrian del Castillo. I know it will take a miracle to get substantive help from any of them, but I still want advice about where I should pin my hopes. Or is HQ effectively saying, once we get past Labor Day, it's everyone for themselves?

[RAY@HQ]

The CDG basically goes useless at this time of year. What you're seeing with the number of positively valued players is basically that there aren't enough stats left in the projection universe to derive dollar values. This will just get worse over the next couple of weeks.

Further, because of the above and due to the need for us to pivot right into producing the 2027 Forecaster immediately after the season ends, as of tonight we're no longer going to be adjusting projected playing time for balance-of-season purposes. Instead, we're going to start staging projected playing time for 2027. So you're going to see guys who are out for the season who have been zeroed out of PT% "coming back to life", etc. So, the CDG in particular and the projections in general are not going to be reliable for the last two weeks. But that's primarily because a two-week data set isn't really usable anyway.