Yup. I agree there should be simple controls to verbosely be able to dictate the exact flows, when we want. So instead, we have to play with the configs to get what we want. I've given up complaining about it (never seen that work anywhere on these forums) and simply try to work around the stuff I don't like - which is still better than any alternative systems I've been able to find out there on the market still.
So for you, is the situation (where it keeps powering the house into the off-peak period after the 9pm peak period) occurring on every single week day, or just Fridays? The other thing to keep in mind, at least in my case, the phone app can be delayed and at times and will show "old" flows that actually ended 5-30 minutes into the past, possibly more, for a few moments or minutes after the phone "app" is loaded. The only way to to know what your real time power flows are for sure is by loading the web interface directly from your local gateway address. I've added a bookmark on my phone to a web address that takes me directly to the local web interface when ever I feel I must know the real situation. I don't trust the app for real time information if I'm about to make a configuration change decision. Regardless, I have seen this behavior you mention after 9pm peak too, as verified via the local gateway web interface....
I've run into what you have but it's only happened on Fridays where it keeps powering the house after my 9pm peak ends It never does this Monday-Thursday when I have my normal week day peaks configured for the "next day". On other days, I have initially thought I was seeing this occur also, but when I checked the local web interface it turned out the phone app (incorrectly) was simply showing me delayed power flows but the system on mon-thursday actually did stop powering the house, which I later verify by looking at the chart and sure enough if did show it stopped right at 9pm mon-thursday even though the realtime flows view of the app (views right at 9:01-9:15pm-ish) seemed to show it still powering the house.
The best I can figure this is only occurring on Friday's for me because I had only off-peak or super-off-peak configured for the weekend. I.E. there are no peaks scheduled after my Friday 9pm peak period ends. The system seems to think since there is no peak for the next two days it can optimize costs (I,e. cost saving mode) by powering the house for a little longer down to my reserve level because it knows it has two days to replenish the PWs where there are no peak periods. This seems like a programming logic bug to me given the costs I've setup having set 0.00 sell values for any of my on or off peaks, and given the round trip Powerwall Charge/discharge inefficiencies of powering the house during off peak when our off peaks costs are so very low. So hopefully they do eventually fix that, but maybe it's intentional. Not sure.
Anyway, my workaround (so far) that works well enough for me is that I setup a "mid-peak" for a few hours around the mid-day time frame on my satuday/sunday weekend schedule. The mid-peak is setup with 0.00 sell value. With this setup the system will charge the PWs all morning, and then will continue charging the PW ( more slowly) with solar left over after powering the house through my "trick" mid-day mid-peak period. It doesn't discharge the PW but does charge the PWs more slowly for this short mid-peak period. And after the mid-day peak period ends, it resumes sending all solar to the PW. This works fine for me because I can still easily get to 100% by the end of saturday since it's only a short mid-peak window, and the house doesn't usually use much power at these times on weekends. This repeats on Sunday, but I'm already at 100% so the powering the house first during my short mid-peak period really doesn't matter because it's generally already powering the house from solar since the PW is at 100%.
Tangentially, TLDR (clearly) Now, having said all this, when house loads and PV generation fluctuate near equilibrium (house load nearly equal to PV generation), power walls flows will jump around at times as the systems fights to control and balance flows to align with desired configuation. At these times it will look like it's powering the house or sending power to the grid when this is not desired, and it probably is for a few moments (minutes?). This is normal in any systems I've ever used or seen tested. The way these systems have to ramp up and down voltages and frequencies to control and balance flows in multiple directions (to/from PVs, PW(s), Grid, House) is usually accomplished via some averaging algorithms. And these algorithms are always a compromise between being too direct and then prone to oscillation, or slower to react yet more stable. I feel like the gateway/PW system strikes a reasonable balance. In addition, during these times it seems clear if a particular balancing "oscillation" event causes a "pull" of power from the grid, when it shouldn't, it will in turn for a moment send a balancing load "back" to the grid to balance the unintentional "pull' from the grid.
One more thing I may try this coming weekend is setting up 0.00 buy AND sell values for all off-peak periods during the week and weekend schedules, and then remove my short mid-peak periods (as a test) for the weekend. If it still powers the house after my 9pm peak period ends (as confirmed via the local web interface, not the phone application) from the PW with an 0.0 off-peak buy and sell values set, on Friday night, then I'm guessing Tesla does have a bug that I hope they will eventually recognize and fix even though I know there is nothing we can do to influence that I'm sure. I can't imagine this behavior would be intentional even if they think they know how to do "cost savings" better than us with manual configurations for our particular situations.