chore(optimizer): bundle Energyplan 0.5.1

[frahlg]

Problem and result

FTW ships Energyplan 0.5.0. On the home box, plans with the car could cost up to about 10 kr more than the best plan within Core's 1.5 s car budget on a Raspberry Pi 4. The cause sat in the planner: once it had met the car's charging goals, a round-off in the goal search made it drop the step that plans the battery around that charging, so the plan kept a battery schedule chosen without regard to cost. A second check threw such plans away for a pricing difference that does not change the plan.

Energyplan 0.5.1 (private source, srcfl/energyplan #63–#72):

• After its goal stages, the planner plans the battery around the car's charging again.
• A car that needs less than two minimum charging runs gets one run at the cheapest time. A need of at most half a run (about 155 Wh at 4.14 kW) counts as met, matching Core's departure-miss rule from docs: let gap_satisfied cover the fleet search too #1532/fix(planner): count a departure miss only above half a minimum run #1533.
• Fixes from Codex's review of 0.5.0: no search bound on a plan the search did not produce, time left for the fallback when the idle-car method stops, cost polish of the car schedule when no joint search follows, complete trace summaries, and EV powers within 1e-5 W of a step sent as that step.

Protocol v1, signs and Core validation are unchanged. Core still validates every plan.

Evidence

Measured on a laptop on 228 recorded battery + car requests from the home box (26 Sep to 3 Oct), each against a reference optimum. On fleet requests the Pi 4 is 17–20 times slower than the laptop, so Core's 1.5 s is about 85 ms there.

• With a fixed search size (4 nodes, so load does not matter): summed gap fell from 112,441 to 21,634 öre; 51 plans cheaper, none dearer, no worse goals.
• At 85 ms: summed gap fell from 148,837 to about 65,000–79,000 öre (runs vary with load).
• The request behind the 10 kr plans (a 253 Wh need due in two hours): 0.5.0 returned a plan costing 924 öre at 85 ms; 0.5.1 returns the optimum, −116.89 öre, at every budget tried.
• These are measurements on one home's requests, not a promise of savings for every site. Long horizons at the 0.5 s budget still leave a gap; that work is tracked privately.

Verification

• make native-solver-test and make verify pass on this commit.
• The bundle is built from private commit 36bd719. Its tree is identical to 1c69e29, which passed full Energyplan local CI (22/22).
• Not yet run: beta on real sites. Merge to master only; do not cut a beta before stable v0.140.1, since a higher beta tag would block that release.

Checklist

• The change follows VISION.md and one selected scope.
• I checked overlapping PRs and coordinated shared files/contracts. No open PR touches optimizer/native or the two Energyplan tests.
• 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 (patch), or the change is exempt.
• 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 2026-10-07T17:27:36.804107Z	e11925e	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.