Thanks for sharing. I would like to offer a counter argument. Perhaps there is no ideal solution, but at least it is worth consideration.
I was particularly excited about V3 because it seemed to be very well constructed for balancing daily load. As V2 only affected review cards, the daily limit had only limitted impact on ebb and flow, but the current V3 had the ability to just about assure a set number of cards per day. If I have set 50 reviews, I would either have my deck limited to 50 cards if there were more reviews, or if there were fewer, new cards would be introduced to reach the limit. In this way it was self controlling.
The most important point in this new behavior (as far as concerns load balancing) is that new cards are not introduced if it means the daily limit would be surpassed. I assume this aspect is not being reverted, so that is a good at least. This introduced a new control mechanism whereby the software could actually self-regulate to adjust for flow. Previously, there was no control mechanism at all: the portion of cards that had review status could be smoothed out, but unless the user carefully adjusts the daily new cards count, the future card load would either continue to increase and pile up or fall below the regular minimum.
The current V3 allows the feedback for this control mechanism (whether or not to add new cards) to be measured by the real daily load on the user. Reverting this particular change separates the control mechanism from the feedback (total count of review cards). In the case of longer learning intervals (which some of us use in order to cope with the ease factor reducing too soon on recently learned cards), this separation could be a few weeks: when a group of new cards are introduced, they will not enter the set of review cards for days or weeks, and so Anki would keep adding new cards because it can’t detect the surge yet. Finally, some of those new cards would eventually make it to the review cards and start to cause the daily limit to be met, shutting off the flow of further new cards. At that point, the learning count would be rather high. Eventually it would work out over a few weeks while no new cards are being introduced for a number of days, but there would be a considerable cycle of ebb and flow.
I have obvserved that learning cards tend to take up the bulk of the study time for my decks where I am actively adding new cards. For example:
Under the old system, the review cards are neatly balanced (but for this recent hiccup where I am preparing to move and my schedule is disrupted), but the learning load is wildly varied.
For this reason, the control of the addition of new cards ideally would be measured by the total actual load rather than considering the review cards only. That does not necessarily mean that the learning cards should be limited as they are in the current V3, but if not, then the mechanism for allowing new cards would have to be decoupled from the daily review limit and instead be tied a separate setting created for something like “Add new cards only when total daily cards are less than [__].”
I understand the needs to balance both programming complexity and complexity for the end user, so maybe there is no good solution here, but thanks for thinking about it anyway.