Repository navigation
Description
What happened
On a 3×16 A site (1 A safety margin, so a 15 A working limit), a car showed "Limited by the main fuse" for about 2½ minutes while the phases read well below the limit. The site has an Easee Cloud charger, a 20 kWh Pixii battery and runs planner_arbitrage on v0.163.0-beta.1. A battery boost (#2012) was active.
Timeline, 2026-11-14 CET. Phase currents come from the site meter in 10 s buckets.
| Time | L1 / L2 / L3 (A) | Battery | Car | EV cap |
|---|---|---|---|---|
| 12:45:30–12:46:30 | 6.5 / 14.8 / 5.6 | +4.2 kW (grid charge from the plan) | 0 | car not asked |
| 12:46:39 | the plan asks for 11 kW | 0 | ||
| 12:46:4x | 2.2 / 9.2 / −1.0 | stops charging (car before battery, #1872) | 0 | 0 |
| 12:46:5x–12:47:2x | 0.9 / 8.3 / −1.0 | 0 | 0 | 0 (35 s dwell) |
| 12:47:24 | start at 6 A (4140 W), the room L2 had | 4140 | ||
| 12:48:1x | 2.5 / 6.1 / −3.2 | −5.9 kW (boost) | 4.05 kW | 4140, fuse_limit |
| 12:49:04–19 | L2 exporting | −6.3 kW | 4830 → 6900 | |
| 12:50:39 | −9.5 kW | 8280 |
The car waited 45 s to start, then sat at 6 A for 100 s with L2 reading 8 A, then 6 A.
Why, as far as I can read it
- Start delay:
evRestartDwell(35 s) before an idle start. - Raise delay: raises read
levelA, the higher of reading and heat, plus pending load. L2's 60 s heat still carried the battery's grid charge. On top of that, the 4140 W start offer counted as unshown for Easee Cloud's 180 s, on top of a reading that already showed the car drawing it.
The start itself is already dynamic. stepEVIdleStart starts at the room there is, not at 16 A.
What I tried, and why it is not a PR
- Raises read
startLevelA, so heat counts only over the line. These fail:TestEVFuseLoadCyclingOnTheOverPhaseDoesNotHuntTestEVHardLimitSixteenAmpPeaksCutOnceAndSettleTestSiteSimNoisyLineCyclesSlowly: "a raise we sent kept a phase over 16.0 A for more than 3 s", and 22 cuts an hour where the limit is 5TestSiteSimScenarios: peak phase 26.7 A
- Skip the dwell when no cut of ours stopped the car (
lim.cutAt.IsZero()). These fail:TestFuseLinePropertyTestEVIdleStartWaitsForTheHeatOfARealOverageTestEVHardLimitRestartsAfterMeterGapOnlyOnSustainedRoomTestJoinedEVWaitsForMeasuredBatteryBackoffThenResumes
Both rules are load-bearing against cycling house loads, meter gaps and real overages.
Proposal to discuss
The room here was freed by FTW itself, not by a house load that may come back:
- The battery stopped a grid charge so the car could start. That order stays off while the car charges, yet its heat keeps the car at the minimum for about a minute.
- A battery boost discharges for the car (fix(control): a battery boost makes the battery cover the boosted car #2012). Every boost tick moves the battery order, so
seeOffersnever reads the charger's landing (batteryOrderMoves). Each raise then waits out the 180 s reservation.
Possible directions:
- Take out of the phase heat the part our own battery orders put there, and count it back if the battery resumes.
- Let a raise's reservation end on the measured landing (
riseAheadA) also while battery orders move, by subtracting the battery's own measured change from the phase move.
Separately, "Limited by the main fuse" during the dwell or while the fuse cools reads to an owner as a fault. A distinct reason would help.
Related: #2012 (boost covers the car), #1948 (Easee start delay).
🤖 Generated with Claude Code
Activity
frahlg commentedon Nov 15, 2026
Thanks for the clear timeline, and for running both ideas against the suite. Here is a proposal.
What the code does today (master):
EVLimitReasonreturns onlyimport_limitorfuse_limit. Core has not setfuse_cooldownsince fix(loadpoint): the site-meter limit is the only fuse guard for chargers #1774, but the box page, the web app and both phone apps still have words for it ("Waiting after a site power limit").readEVDrawandnoteDrawalready keep the charger's own samples with their times, but only while no order of the run has surely landed (foundOpen).Proposal, in three steps:
fuse_limitonly when the reading binds, andfuse_cooldownfor the rest. The clients already show it, so they need no change.evUnknownSettle(180 s for Easee Cloud) holds as today. This measures an existing rule better; it adds no rule. It does not wait onseeOffers, so a battery order that moves on every boost tick no longer stalls it. Evidence: every fuse test and the site simulation pass, plus a new scenario that replays this issue (battery grid-charges, the plan asks for the car, a boost covers the car).Not recommended: taking the battery's share out of the phase heat. It needs the battery's current per phase, which Core does not know for a single-phase battery on an unknown phase. It adds a rule to fuse protection. And the heat in the fuse is real, whatever caused it.
Today's cost is a couple of minutes of slower charging per switch. That is not a safety problem; the words "main fuse" are what make it look like a fault, so step 1 matters most to owners.
You have the site and the context, so you are the right person to build steps 1 and 2, as one PR or two.