Sezibwa SRNE inverter: proven behaviour register

Copy of the register kept in the microgrid-gateway repo: docs/sezibwa-srne-proven-behaviour.md . The repo copy is the master; this page is refreshed when it changes. Owner: Hillary (with Claude). Reviewer: Aaron. 
 
 Site: Sezibwa Rentals, gateway gw-pi2. Inverter: SRNE ASP48120SH3, with 1 battery and 8 × 600 W panels. 
 This register lists what we have seen happen on this unit, with the date and the evidence. A control rule may depend only on rows under Proven . Anything under Not proven needs a planned test, with a clear pass and fail, before a rule relies on it. 
 Rules for keeping it 
 
 Add a row only when the data shows it, and cite the evidence (Influx field and time, driver log line, or register read). 
 A result counts only if no other cause can explain it. On 11 Nov the charge stopped near 50% because the 00:40 window ended, not because of E03B. That case did not separate the two causes. 
 When a rule changes, its PR lists the rows it relies on. 
 
 Times are EAT (Africa/Kampala). 
 Proven 
 
 
 
 # 
 Behaviour 
 Date 
 Evidence 
 
 
 
 
 P1 
 The output mode changes only when E204 is written; the manual's implied mode is not applied 
 3 Nov 
 Stayed SUB until E204=2 was written (v29) 
 
 
 P2 
 SBU runs the houses from sun plus battery, with the grid unused, when SoC is at or above E020 (after a drop to E01F the unit stays on the line until SoC reaches E020, see P4/P5). Every SBU observation so far was at or above E020; below it is N5 
 4–5 Nov, 12 Nov 16:00 
 MachineState 3, meter shows no grid draw 
 
 
 P3 
 In a grid cut the unit runs on battery, and stays on battery after the grid returns until the battery reaches E01F 
 6 Nov 00:37, 11 Nov 
 SoC timeline; stayed in state 3 with the grid at 230 V 
 
 
 P4 
 At E01F with the grid present, it moves the houses to the grid (state 2) 
 11 Nov 23:05, 12 Nov 22:15 
 MachineState 2 at 10–12% SoC 
 
 
 P5 
 With E20F = 2 (hybrid), the grid charges the battery at about 120 A at any hour while the unit is on the line 
 11 Nov 23:05–23:50 (to E020 = 50%); 12 Nov 14:00–16:00 (SUB, solar near 0, 116 A) 
 BatteryCurrent −117 to −120 A, solar about 0 
 
 
 P6 
 The timed charge (E00E = 1, window E026/E027) charges from the grid at about 120 A, about 1% SoC per minute 
 13 Nov 00:05–01:50 
 SoC 11 → 99 % 
 
 
 P7 
 E03B (window stop SoC) does not end the timed charge 
 13 Nov 
 E03B = 50, charged to 99% inside a 00:00–06:00 window 
 
 
 P8 
 The solar panels keep charging while the unit is on battery during a cut 
 6 Nov 10:22 
 PV 1.4–2.0 kW, battery charging at +0.7–1.3 kW 
 
 
 P9 
 In SUB with the battery near full, the solar input is shut down (PV voltage 0) 
 10 Nov (94%) 
 Pv1Voltage 0 at midday 
 
 
 P10 
 The panels can reach about 98% of nameplate (4.7 kW of 4.8 kWp), so the panels and wiring are fine 
 11 Nov 12:50 
 charger0/Pv1Power max 
 
 
 P11 
 Settings are written only in MachineState 2, except the ModeSchedule SUB write, which is allowed in state 3 when the grid is present on all phases (#45, gridPresentMinVoltage 180), and the E20F safe value (#48, not yet deployed) 
 design, logs, Influx 
 Writes stayed queued all day on 11 Nov while the grid was off. State-3 SUB write: 12 Nov 14:00 EAT, MachineState 3 / OutputPriority 2 at 13:40–14:00, then state 2 / priority 3 at 14:00–14:20 with the grid at 234 V (Influx, 20-min means; driver log lost when the container was recreated on 13 Nov) 
 
 
 P12 
 A charge-window change can stall after the disarm or window write; a controller restart while in state 2 completes it 
 11 Nov 18:11, 12 Nov 14:00 
 Log shows the disarm or window write with no follow-up; after the 14:31 restart, enable = 1 at 14:34 
 
 
 P13 
 ModeSchedule times are Kampala local time, not UTC 
 10 Nov 
 v35b fix 
 
 
 P14 
 At E01F on the line, the unit sometimes does not grid-charge (stays at about 0 A) 
 12 Nov 22:15–24:00 
 SoC 11%, BatteryCurrent about 0, state 2. Compare with P5; see N2 
 
 
 
 Not proven (test before relying on it) 
 
 
 
 # 
 Question 
 Why it matters 
 Planned test 
 
 
 
 
 N1 
 Does E20F = 3 (PV only) stop all grid charging, including the timed window? Is E20F re-read during an active timer? 
 D138 (openems #48) depends on it 
 First night after #48: log E20F, the window and BatteryCurrent from 23:50 to 00:30 and at any E01F transfer. Pass: no grid charge outside 00:00–06:00, and the night charge runs 
 
 
 N2 
 Why did P5 happen on 11 Nov but not on 12 Nov (P14)? 
 Predicting evening grid buying 
 Compare the registers (E20F, E205, E020, OutputPriority) logged on both nights 
 
 
 N3 
 Cause of the 5 Nov solar stall (PV fell from 2.8 kW to under 0.7 kW on the switch to battery) 
 Daytime SBU switching 
 Watch every daytime SBU switch for a PV drop greater than 50% 
 
 
 N5 
 Does an SBU write below E020 move the unit to battery, or does it stay on the line until SoC reaches E020? 
 #114 (daytime SBU from 30%) depends on it 
 After the first daytime SBU write below E020, watch MachineState for 5 minutes. Pass: state 3 within a minute 
 
 
 N4 
 Real daily solar compared with PVGIS (about 18 kWh/day expected) 
 The business case 
 One full month of charger0/PvEnergyToday against PVGIS for the same days 
 
 
 
 Change log 
 
 13 Nov 2026: register created from 3–13 Nov site data. 
 13 Nov 2026 (Aaron's review): P2 limited to SoC at or above E020; P11 lists the state-3 exceptions, with evidence; N5 added.