fix: LOAD_SUPPORT discharge ceiling leaks real-time load spikes to grid despite ample SOE headroom

[johanzander]

Context

Split out from #381 after diagnosis. Frank-Leysen (Growatt MOD 5000TL3-XH via solax_modbus_growatt_min, TOU mode) reported small recurring overnight grid imports (0.1-0.2 kWh, a few tens of cents/night) while SOE had 3+ kWh of headroom above min_soe_kwh. Confirmed against his bess-debug-2026-07-21 / bess-debug-2026-07-22 bundles: every import happens in the same 15-minute period the battery is actively discharging under LOAD_SUPPORT, with the discharge-rate register at exactly the ceiling written for that period.

Root cause

_map_intent_to_rates (core/bess/inverter_controller.py:479-486) uses one shared branch for both LOAD_SUPPORT and BATTERY_EXPORT, capping the discharge-rate register at the DP's planned average power for the period:

elif intent in ("LOAD_SUPPORT", "BATTERY_EXPORT"):
    if battery_action_kw < -0.01:
        discharge_rate = self._scale_to_percent(abs(battery_action_kw), self.max_discharge_power_kw)
    else:
        discharge_rate = 0

These two intents are not physically equivalent:

• BATTERY_EXPORT → grid_first: no deficit backstop. An open ceiling directly oversells stored energy beyond the arbitrage plan for the full period, regardless of load. Capping this is correct and should not change.
• LOAD_SUPPORT → load_first: self-limiting to the real house deficit (inverter_simulator.py:132-143, delivered_kwh = min(deficit, rate_kw*dt, ...)). A load spike above the written ceiling leaks straight to grid even with ample SOE and physical discharge capacity (max_discharge_power_kw) sitting unused, because the register — not the deficit — is the binding constraint.

Note: LOAD_SUPPORT is deliberately not hardcoded to 100% — that was the bug fixed in #147 (battery drained early on high-consumption days, blowing past reserve planned for a later, pricier period). So the fix is not "remove the cap," it's making the cap economically aware.

Proposed fix

Add a shadow-price-gated discharge ceiling for LOAD_SUPPORT, mirroring the existing SOLAR_EXPORT shadow-price gate (docs/agents/bess-knowledge.md "Governing Economic Law" section, battery_system_manager.py solar_export_discharge_rate()): allow the discharge ceiling to open above the plan-scaled value specifically when the DP's own shadow price for that period indicates the marginal stored kWh is not earmarked for a later, more valuable use (i.e., genuine surplus headroom exists). Keep BATTERY_EXPORT's ceiling exactly as-is (plan-scaled, no gate) since it lacks the physical backstop that makes this safe for LOAD_SUPPORT.

Acceptance criteria

• LOAD_SUPPORT discharge ceiling is no longer a single shared branch with BATTERY_EXPORT.
• When shadow price indicates no future need for reserve, LOAD_SUPPORT can open beyond the plan-scaled ceiling to cover a real-time load deficit up to max_discharge_power_kw.
• When shadow price indicates the reserve is needed for a later period, current plan-scaled behavior is preserved (no Battery discharge cannot be paced for home support (LOAD_SUPPORT dumps greedily on deep evening peaks) #147 regression).
• Regression test replaying Frank's bess-debug-2026-07-21/-22 bundles shows the previously-leaked overnight import periods now served from battery instead of grid, with no change to daytime/evening schedule.
• BATTERY_EXPORT behavior is unchanged.

Related

• feature: computed overnight reserve floor (Min SOC + historical night use) so one knob covers evening export + night load #381 (original floor proposal — superseded by this diagnosis, see comment history)
• Battery discharge cannot be paced for home support (LOAD_SUPPORT dumps greedily on deep evening peaks) #147 (original LOAD_SUPPORT pacing fix — do not regress)
• docs/agents/bess-knowledge.md SOLAR_EXPORT shadow-price gate (design precedent)