[frahlg]

Why

Tonight's beta carries #1828 (an idle car starts on the hour's slack under the import ceiling) and #1814 (interchangeable batteries share the plan's total). The site pack had no case for either. Its house also plans with Core DP, which refuses two batteries, while a beta box plans with Energyplan.

Test-only: go/test/site. No Core change, no changeset.

What

• s16-25, import ceiling (the help report behind fix(control): an idle car starts on the hour's slack under the import ceiling #1828). 25 A, house 1.3 kW, the battery held idle at its floor, POST /api/peak_import_ceiling 4 500 W, Charge now 16 A. House plus the car's 4 140 W minimum is above the ceiling. Core counts the part of an hour before its first reading at the ceiling, so the case starts in an hour's first 4 minutes or waits for :56, then runs one whole clock hour and 3 minutes. Criteria:
  • the car starts on the hour's room before :45;
  • every clock hour's recorded import averages at most the ceiling + 1 %;
  • the next hour stops the car within 60 s;
  • while held, it says import_limit in 90 % of samples and never fuse_limit;
  • the fuse holds.
• s17-25, two equal batteries (Rann's site behind fix(control): interchangeable batteries share the plan's total #1814). Two hybrids with 10 kWh at 74 % and 60 %, house 3 kW, no PV, Energyplan with a 6 h horizon. Energyplan puts each quarter's discharge on one battery, as on Rann's site. Over its usual 48 h it keeps both batteries for the dearest hours. Today that was Monday evening at 40–60 öre, so the current quarter stayed idle and Core's per-battery path never ran. Over 6 h the batteries hold more than the house needs, so the plan discharges now. The case still waits up to an hour for a quarter the plan discharges, then runs 15 minutes. Every 30 s it saves the plan's order for each battery in plan.jsonl. Criteria:
  • Energyplan's plan moves the batteries in half the polls. This is the setup check, tolerated: the day's prices decide;
  • while the fleet moves 500 W or more, each battery takes at least a quarter of it 90 % of the time;
  • while the plan moves them, the charge gap shrinks by more than 0.2 points;
  • the fuse holds.
• FTW_SITE_ENGINE=core|energyplan picks the planner for a whole run. Default core, as before. For energyplan the run directory links optimizer/native/bundle, where FTW looks for the worker. Only a run with Energyplan gets the link, so Core DP runs keep today's behaviour.
• The house record now keeps phase_w (site power) and a second battery (inverter2). /api/loadpoints reads manual.limit_reason.

Evidence

All runs: device-simulator a3030d3, drivers 219c462e, on the M5 laptop. Each Core is built from the commit named.

Case	Core	Verdict	What the record shows
s16	master 69700f41	pass	Car starts 12:24:39, 1 479 s into the hour. Hour means: 12:00 3 982 W against 4 500 W; 11:00 1 300 W and 13:00 1 351 W (partial hours). Stops 2 s after 13:00. 1 415 held seconds, every one import_limit, none fuse_limit. Worst phase 8.5 A.
s16	0c8eaef9, before #1828	fail: the car never starts	Hour 13:00 mean 1 300 W, the house alone; 3 377 of 3 387 held seconds say import_limit. This is the help report.
s17	master 69700f41	pass	The plan moves the batteries in 31 of 31 polls, all on one battery (−3 000 W / 0 W). Core gives 1 684 / 1 341 W, which is 69:55, the energy each holds above 5 %. Gap 14.0 → 13.1 points.
s17	1883582e, before #1814	fail: each battery takes a quarter	3 015 / 0 W for all 900 s, as on Rann's site. The gap shrank here too (13.6 → 6.1), because the plan happened to name the fuller battery; only the share catches the fault.
Full set without s16 and s17, FTW_SITE_ENGINE=energyplan	master 69700f41	24 pass, 4 known, 0 fail	The known four are s2, s5, s5s and s13, the same four as Core DP on 16c4ad09. 85 min in three lanes.

go vet ./test/site/ and go test ./test/site/ (the unit tests; TestSite skips without FTW_SITE) pass.

Timing: s16 runs about 65 min, longest first in its lane. Started between :04 and :56 it first waits for :56. s17 runs 17 min when the plan discharges at once, and up to 77 min when it does not.

Not here

• A duty quarter that crosses an hour boundary near the ceiling (fix(loadpoint): a duty quarter runs the car for the plan's on-time #1827). The house runs at exactly 230 V, so a quarter's energy never runs out early. fix(loadpoint): a duty quarter runs the car for the plan's on-time #1827's own TestSiteSimDutyOnTimeYieldsToTheImportCeiling covers it.
• The runner still runs Core DP. Switching its full set to Energyplan is a separate choice.

🤖 Generated with Claude Code