Ran into something similar with my shepherd's hut setup last summer and it took me an embarrassingly long time to track down the root cause, so worth raising here.
When Victron ESS starts exporting rather than charging, the first place I'd look is the DVCC charge current limits. If your BMS is sending a CCL (charge current limit) of zero — even momentarily — the Multi will flip its behaviour and the ESS algorithm can interpret that as "battery full, export excess." The JK BMS in particular can be a bit trigger-happy with its protection flags.
A few things worth checking:
- VE.Bus BMS vs shared current control — if DVCC is enabled but the BMS CANBUS integration isn't properly configured, you can get conflicting signals
- Grid setpoint — worth confirming your feed-in/setpoint values in ESS assistant aren't inadvertently set to export preference
- AC-coupled PV on AC-out vs AC-in — makes a significant difference to how the ESS assistant handles power flow priority; Huawei inverters on AC-in behave differently to AC-out configurations
- Check your Venus OS logs during the problem period rather than just the live dashboard
On my boat I had a vaguely similar issue where the Pylontech BMS was throttling charge current due to cell imbalance, and the system just quietly decided exporting was preferable. Took a VRM deep-dive to spot it.
What does your VRM data show for the BMS charge current limit at the time it starts exporting? That's usually the smoking gun. Would be interested to know what firmware your JK units are running too.