feat(planner): plan with the published prices only

[frahlg]

Closes #1439.

Why

Owner decision: the planner uses the prices that are published. The forecast filled every slot after the last day-ahead price with an hour-of-week climatology at confidence 0.6. It fed invented prices into charging decisions and into the terminal value. #1154 shows the tail even used a different VAT. Published prices always reach 11–35 h ahead, and the nearest hours carry the value.

Change

Core

• Deleted go/internal/priceforecast: the model, the 6 h refit loop and the <config dir>/seed/prices.csv import. No such file ships.
• mpc: removed PricePredictor, Service.Price, extendPricesWithForecast and Service.GridTariffOreKwh, which only the forecast used. VATPercent stays for demand charges.
  • The plan ends at the last published price. Horizon (48 h) stays as an upper bound.
  • mpc: replanned now logs prices_until.
• Removed Confidence from slots, plan actions, diagnostics and the Home Assistant plan and price attributes. With it went the DP's blend toward the horizon mean and the "(predicted)" reason tag.
  • horizonMeans became horizonMeanPrice, since the export mean only fed the blend.
• Energyplan. The request still requires confidence (deny_unknown_fields → missing field 'confidence' without it), so Core sends a constant 1.
  • Removing it from the worker needs an Energyplan release and a macOS build, for no change in behaviour.
  • Tracked in srcfl/energyplan#41 for the next release made for another reason. FTW's native-solver tests fail loudly if the two drift apart.
• pricefc/state stays in the config table with a note in migrate(), as with Home Link and Ask why. Nothing reads it, and an older release restores its model from it after a rollback.

Web

• Plan view: no predicted zone, predicted bars, tooltip badge or "Predicted (ML)" legend.
• The brief's Forecast field says Published prices · Plan until Sat 24:00. A plan ending at midnight is named by its own day.
• The chart's default view ends where the prices end, capped at 48 h.
• The Plan help and the Settings › Planner help say the plan reaches as far as prices are published.
• Diagnose: the Conf column and dimming are removed.

Behaviour afterwards

• A replan covers the published prices. Here, at 14:11 with tomorrow's prices out, that was slots=136 … prices_until=2026-09-27T00:00.
• The terminal SoC value is computed over the published prices.
• An EV deadline later than the last published price is clamped to the plan's last slot, as before. Before about 13:00, a target for tomorrow morning is met within today's known prices until the replan after publication moves it.

Evidence

• make test: all 67 Go packages ok.
• make native-solver-test against the bundled Energyplan 0.4.8 worker passes. Without confidence, every native test failed with missing field 'confidence', which is why Core keeps sending it.
• npm test with LANG=C: 639/639. A Swedish locale fails update-dialog.test.mjs on master too, because of "8 192".
• Browser. Local Core on this branch, with elprisetjustnu SE3 prices and a simulated Sungrow:
  • brief: "Published prices | Plan until Sat 24:00";
  • legend without "Predicted";
  • the tooltip has no predicted badge;
  • no JS errors;
  • the chart ends at the last published price.

Not in scope

PV and load forecasting is a different "forecast" and is unchanged: forecast, forecasting, planner.forecast_trust, the PV safety k slider and plan-forecast.js.

Notes

• fix(prices): unset VAT is 25% in the store and 0% in the planner forecast #1154: this removes its forecast-tail symptom. The VAT default split remains: the demand charge uses raw vat_percent, and true 0 % VAT cannot be set. It stays open.
• fix(web): pause plan, heating, settings, and card polls when hidden #1177 (hidden-tab polls) also edits web/plan.js, in its polling code. The hunks here touch only drawing and the chart window, so conflicts should be small.
• Changeset: minor, because the plan horizon and the Plan view change visibly. The owner requested the change.

🤖 Generated with Claude Code

https://claude.ai/code/session_01MuerPFZFG88kgu8sWVHeq7

[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 2026-09-25T12:16:51.498738Z	66b7fe7	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.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 66b7fe7d74

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".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Update the MQTT payload documentation after removing confidence

Removing confidence from both schedule payload structs makes the maintained integration guide incorrect: docs/ha-integration.md lines 49–55 still promises confidence in plan_json and describes plan_schedule_json as containing every field. Home Assistant or raw MQTT consumers following that contract will wait for a field that is no longer emitted, so update the documented payload alongside this wire-format change.

AGENTS.md reference: AGENTS.md:L321-L322

Useful? React with 👍 / 👎.