Someone wrote in the forum about a seemingly small glitch of the training center which under the surface is a pretty complex topic which kept me busy quite some time while I worked on the design of CPT 4. I have to admit this is a rather dry topic if you are not into it 
Background info
First you might want to read his post.
In short: If you train with the option "always jump to worst position" it is possible that CPT will present you positions which are not reachable by a top-ranked candidate move in your sub-repertoire. This is a training setting for advanced players which know their repertoire pretty good. Basically the difference is that you will not be lead to those positions from the start position on, but instead just presented the final position - the currently worst position of your sub-repertoire. A great way to really recognize a position and not move orders.
I'm aware of this limitation for quite some time. Unfortunately it is not
so easy to improve. Actually a similiar issue exists for training of
tactical positions or the calculation of the sub-repertoire overall
training score. Here again CPT should ignore those positions which are not reachable by top-ranked candidate moves or in the case of training of tactical positions CPT should pick the start position and not any / worst of the (usually short) line. The latter has some additional challenges which are not covered by this post, but by CPT 4 
The Problem
At the moment CPT can't tell whether a certain position is only reachable by a non-top ranked candidate move or not.
Two possible approaches
There
are two approaches to solve the problem. Either calculate this
information "on-the-fly" when CPT needs it (for example during the
training session) or pre-calculate it for each position and save the
info.
Believe me, for CPT 4 I spent hours, well, overall days
about just this topic and how to design and develop the "perfect"
solution. For several times I thought I had covered all scenarios and I
was pretty sure I found the solution - just to find out - no, not yet. I
became almost mad.
OK, let's have a look at both possible approaches to this challenge.
Calculation on the fly
You
quickly run into a performance issue. First you would just calculate
the worst position. Then you would need to check whether the position
is reachable by a top-ranked candidate move / line. If you have more
and more of such non-top ranked positions over time the performance
will (significant) decrease (just imagine to check 1000 positions and
their preceding moves before you finally find a "right" one...). There
is even a bad side-effect that those non relevant positions will be
always checked first as they are never trained and remain at a training
score of 0%. I don't believe in this approach.
Pre-Calculating
Obviously
this solves the performance issue. However, now things get really
tricky if you start to think about the details. I encountered the same
concepual issues as I did with the idea of an extended (sub-)repertoire
concept.
You will need to check for each new move you add to the
repertoire whether the new arising position is now a valid position for
training or not (or for the repertoire training score or a
sub-repertoire). So far so good. Thanks god there exist transposition
in chess.
Imagine to reach an already existing position which was
so far NOT a valid position for training. However, now it is reachable
by a top-ranked candidate move. The logical consequence would be to
change the flag for the position and all consecutively positions which
are reachable from that position as "valid". Vice-versa if you delete a
move. Or think about what should happen if you manually change the order and thus the rank of a candidate move...
Also consider more complex actions like an import of
thousands of games or a copy of a sub-repertoire with >10.000
positions. In these cases CPT would need to check for each position
whether it is a valid position for training or by transposition any
already existing position (and their consecutive positions) should
change their flag now. This will lead to a huge performance penalty for all these actions.
Maybe I'm just too pessimistic (I'm usually not), but I believe
there are good chances that somehow this information gets messed up.
Then a non valid position will be marked as valid and vice-versa. Good
chances for "chaos" here.
New approach
CPT 4 will not care for the info "valid" position for training or
not while you add a new move / position (either by manual editing, import or copy / paste). Instead it will offer you a function to start a batch job which
just goes through all branches and makes sure the flag
is correctly set for each position. Take into account that you only have to run it once you have added a new move for your side which is not a top-ranked candidate move. If you always only add one candidate move for your side and many for the opponent you don't have to care at all!
I don't like the aspect that you
have to manually start this job as there are good chances some users
will miss it. However, those will likely not bother anyway and I
believe it is still - after consideration of all CON's and PRO's - the
best approach. Sometimes there just doesn't exist the goose that lays
golden eggs.
Well, who knows. This apparantly simple and minor issue cost me so
many hours and sleep, maybe I will change my mind once again 
Cheers,
Stefan