[frahlg]

Problem and result

A beta user's help report (25 October 2026) has the hourly import ceiling on: peak_import_ceiling_w 4,500 W, hourly-average mode. The car's minimum is 4,140 W and the house used about 0.85–1.3 kW. For a whole hour the car stayed at 0 W (fuse_cap_w=0, reason import_limit), although the hour had 3.3–3.9 kWh of room left.

The ceiling caps the clock hour's mean import, but an idle car could start only if house load plus its minimum fit under the ceiling at that moment. Here 0.85–1.3 kW + 4.14 kW is above 4.5 kW, so the car never started. A car that was already charging would have been allowed to keep going under the hour's rate and rise rules.

Now the idle car starts once the hour has saved enough room. It charges to the end of the hour, the next hour starts again at the ceiling and stops it, and every hour's mean stays under the ceiling. Owner request: "fix all bugs".

Cause. go/internal/control/ev_fuse_budget.go:575 (master) set r.startLevelW = math.Min(riseW, s.PeakImportCeilingW). stepEVIdleStart (:532–536) needs startLevelW − dwell's mean import − pending ≥ minimum, so for a start the hourly average worked as an instant cap. The rule came with #1699 ("on the hour's slack alone it would start at about :35 and stop at :00, once an hour"), and TestImportCeilingNoStartOnSlackAlone pinned it.

Change (ev_fuse_budget.go:573–579):

• The start level is now the rise level, riseW. The rise level already follows the hour's rules (stepHourImportLimits, import_hour.go:149–188). It starts each hour at the ceiling, rises only in whole steps with one step in hand, waits 5 minutes after a cut and never rises in the hour's last 15 minutes. The rate cut still guards the hour's energy.
• In the hour's last 15 minutes (importHourRiseMinLeft) the start level stays at most the ceiling, as before. A start on slack then would run only minutes before the next hour stops it. A car that fits under the ceiling itself still starts then.
• Two comments now describe the new rule. No other code changes.

Scope and safety

• Hour's mean. riseW ≤ limitW always holds, and the cut on the hour's rate is unchanged, so a start cannot overrun the hour. A Core restart, a new ceiling or a new hour puts the rise level back at the ceiling. A meter gap never raises it. Before the hour is measured, the start test does not read the start level.
• Stale data, failure and recovery: unchanged. Stale site-meter data still stops dispatch.
• Labels. While the ceiling holds the car, it says import_limit, whether idle or charging below its want. When the fuse leaves less room, it says fuse_limit, as before.
• Overlap. Checked open PRs touching ev_fuse_budget.go and import_hour*.go:
  • fix(control): fuse protection as one function #1789 (draft, conflicts with master) rewrites stepEVIdleStart and the evMeterRoom constructor. It keeps the old startLevelW line and comment. On a rebase, keep this PR's start level and its last-quarter rule.
  • fix(control): phase current takes its direction from the phase's power #1782 (draft, conflicts with master) changes nearby evMeterRoom field comments, not the start level.
  • fix(charging): carry two-phase cars through plans, current limits and Easee #1792 changes offerPhaseA and evPoolFor only.
  • fix(loadpoint): a duty quarter runs the car for the plan's on-time #1827 (EV keeps its order at quarter end) changes loadpoint/controller.go and sitesim_test.go, not this file.
  • No registry names change.

Limits

• When house load plus the car's minimum is above the ceiling, the charger starts and stops about once an hour. It starts around :17–:32 in the tests and stops at :00. That wears the contactor, and some cars dislike pauses. It is the price of using the hour's room.
• The rise rule is cautious. In the help-report case each hour ends about 0.5 kW under the ceiling (means 3,948–4,024 W against 4,500 W).
• The planner does not see the ceiling yet. Its plan may still expect charging the ceiling does not allow, or the reverse. That is a separate follow-up.
• After the stop at :00, a fuse still warm from a run near its limit can make the idle car say fuse_limit for about 6 s. The label names whichever room is smaller, and the fuse's heat model briefly leaves less. TestImportCeilingNightNeverSaysFuse checks the label from a minute after the stop.
• web/app.js:3001 says "this hour's import budget is used" for an idle import_limit. That is now also shown while the car waits for the hour to save room. The wording is not changed here, because a UI change needs a browser review. Follow-up.
• The site simulation's wantsCharge (sitesim_test.go:400–405) expects a start once house + minimum fits the hour's rate. Core's rise level trails that rate by one or two steps. No current row combines a ceiling with house + minimum above it, so nothing fails. A future row would need that slack.
• No hardware run. Everything below is simulated.

Verification

go/internal/control/import_hour_test.go, the hourSite harness: an Easee-like 3-phase charger (5 s command delay, 10 s late telemetry), a 25 A site, a 1 s site meter, 4,140 W minimum.

Case	master	this PR
4.5 kW ceiling, 850 W house, 09:00–12:00	car 0 Wh; hour means 850 W	starts 09:17:31, 10:17:00, 11:17:00; stops only at :00; about 9.5 kWh to the car; hour means 3,993 / 4,024 / 4,024 W
same, 1,300 W house	car 0 Wh; hour means 1,300 W	starts at :24 each hour; about 8.0 kWh; hour means 3,948 / 3,979 / 3,979 W
7.5 kW ceiling, 5 kW house (the old #1699 case)	never starts	starts at :32 each hour at 4,140 W; hour means 6,918 / 6,921 / 6,921 W
7.5 kW ceiling, 40 min at 500 W, car plugged in at 22:40 beside 6 kW / 8 kW	idle to 23:00	runs from 22:40:01 at 9,660 / 7,590 W, held by the fuse margin; 22:00 hour means 5,536 / 5,515 W. At 23:00 the new hour stops it. 6 kW: restarts 23:42:14, 23:00 hour mean 7,234 W. 8 kW: the house alone is over the ceiling, so it stays idle. From 23:01 to midnight no tick says fuse_limit.
7.5 kW ceiling, 46 quiet minutes (rise level above the fuse), car plugged in at 21:46 beside 8 kW	idle	idle to 22:00 (last-quarter rule), every tick import_limit

Tests:

• TestImportCeilingNoStartOnSlackAlone becomes TestImportCeilingStartsOnHoursSlack, with the first three rows. Over three hours, in every clock hour:
  • the car starts at least once;
  • it stops at most once, and never within 15 minutes of a start;
  • the hour's mean holds the ceiling;
  • every tick says import_limit.
• TestImportCeilingNightNeverSaysFuse is the fourth row.
• TestImportCeilingStartTestSaysImportLimit is the fifth row and pins the last-quarter rule.

Mutation checks:

• With master's math.Min(riseW, ceiling), TestImportCeilingStartsOnHoursSlack (all three cases) and TestImportCeilingNightNeverSaysFuse fail.
• Without the last-quarter rule, TestImportCeilingStartTestSaysImportLimit fails: the car starts at 21:46:01.

Runs:

• go test ./internal/... ./cmd/ftw/... -count=1 passes, with drivers fetched by make drivers. go vet ./internal/... ./cmd/ftw/... is clean.
• go test -race ./internal/control/ -count=1 passes.
• go test ./internal/control/ -run TestSiteSimFieldMatrix: 340 cases, 29 fail as known (idle 23, own-raise 6). testdata/sitesim_field_known.txt is unchanged. No matrix row sets the ceiling, so this change cannot move it. The ceiling scenario in sitesim_test.go ("end of a heavy hour, low phases") still passes.
• An independent review (separate agent) traced the hour's mean through restarts with and without SeedHourImport, meter gaps, a ceiling changed mid-hour, zones and battery pending orders, and found no bug. Its random-load runs (house spikes, Core restarts, gaps; 80 and 96 car-hours at 4.5 and 7.5 kW) had no hour above the higher of the ceiling and the house-only mean. Its wording points are fixed in the changeset and the commit message. The rest are listed under Limits.

Checklist

• The change follows VISION.md and one selected scope.
• I checked overlapping PRs and coordinated shared files/contracts (see Overlap).
• Relevant checks cover the changed behaviour and failure paths.
• A human reviewed changed UI in a browser, or no UI changed (no UI changed).
• A Changeset is included, or the change is exempt (patch).
• Every commit has a DCO sign-off.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-25T09:06:54.332508Z	09dfa52	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.