[frahlg]

Fredrik asked that VISION.md say what FTW is building: a hard, simple core that people, automations and AI agents can build on, chargers included, with forecasting and planning as layers that do not complicate it.

Change

• The opening paragraph names FTW Core as the primitive. The Energy Planner, forecasts and analysis become layers on top, as Home Assistant and MQTT already were.
• A hard, simple core: four things the core does (read, protect, act, record). Readings are shown as they are; Home use is blank only without a fresh site meter. Fuse protection is one physical calculation per phase on the current after a change, counted with its direction, with commands treated as made until the device can respond. The hourly import ceiling has its own name. "Charge now" is not stopped by a guessed SoC, a plan or an unnamed car. A rule belongs in the core only if it protects against a quantified physical risk or carries out a request. A simulated house with a real fuse is the core's test; it is marked as a target.
• Layers on top of the core: layers read and propose, the core decides, and a layer never holds up the control loop. Three streams: live readings as reported, history with a quality stamp per interval, and training data made only of fully measured intervals.

Why now

A review of master on 21 October found nine separate fuse-limit calculations, several confirmed false limits (export counted as import, phase mode decided two ways, the hourly ceiling reported as the fuse) and one path where a slow cloud charger can exceed the fuse. It also found forecast and history logic on the control tick. Each line above answers one of those findings.

#1292 also edits VISION.md, in a different section.

🤖 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 2026-10-21T08:06:32.753851Z	a1455c0	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: a1455c0bcf

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]

Scope the fully measured restriction to forecast training

The absolute “nothing else” restriction conflicts with the battery/inverter twin requirements later in this file: VISION.md:190-193 explicitly permits quality-labeled device-only samples, and the existing battery model learns from commanded and device-reported power on each control tick (go/cmd/ftw/main.go:2959), not from fully measured history intervals. Following this new requirement would discard valid response and saturation learning whenever independent site confirmation is unavailable; limit it to household/PV forecast training or preserve the documented device-only tier.

Useful? React with 👍 / 👎.