goodwe-addon/goodwe_controller/DECISIONS/UX-54-forecast-tab.md

UX-54 - a Forecast tab: the sun, the house and each battery's level, per hour

Status: DONE 2026-10-26 (owner go 2026-10-26: "Day curve", "New tab").

Decision

A tab beside Home and Batteries shows today and tomorrow, per hour: the solar forecast, the solar measured so far, the house's expected use and each battery's expected level with its blocks. It follows the add-on, as Home does (UX-53), and decides nothing.

• Blocks are Home's. Today's are the runs of the same plan Home draws (grid, give, monthly cycle). Tomorrow's come from the same builders (homeview.runs_for()). The words are the plan lines'. A block is on the tab exactly where Home has one, so the UX-53 gate is followed too.
• Tomorrow is read ahead. Home sizes a block from the battery's level now and draws none when there is no room or energy (room() and plan_need() in homeview.mode_runs(), the leg sizing in trade_runs()). For a day that has not started that level says nothing, so tomorrow's blocks are built with ahead (Home never passes it): mode_runs() reads each block at the level the curve has when it starts, with its own filter and MIN_KWH, so there is no give block where the battery is at the floor and no grid block where it is at its planned level. Where the curve has no level yet, a charge is read from the floor and a give from the ceiling. The blocks and the curve follow each other and are taken to a fixed point. The gate (mode_gate, UX-53) is not bypassed.
• Modes off has no curve (no forecast is read, no house profile kept), so trade_runs(ahead=f) reads each leg at Home's own count (dayview._counted()): the level now, plus what the blocks before it put in, less what they take out. No sun is read and the house only takes, so no give block where the emptiest battery is at the floor by that count, and control does not give there. A charge is read from the floor: a full battery covers the house before the cheap hour, and control charges it back then. Walked every 15 minutes on a cloudy day, the packs moved by control and a 400 W house (test_ux54.py part 7, cases A-D, one price quarter per leg), control charges and gives in exactly the quarters the tab has a block.
• A block lasts as long as control acts. A cheap run is sized for all the batteries charging together, so a slower one is still below modetrade.charge_to() when it ends, and control plans the next cheap slots (Controller._price_runs()). dayview._levels() asks the same functions, modetrade.energy() and priceruns.plan(), with the packs at the curve's level, and draws the grid block to where control stops. Each battery's block then ends where the curve has it at its own level, and a give block where it reaches the floor: the curve puts a point at both (daycurve.curve() cuts the step there), and the block ends at the first point where the level the tab shows (0.1 %) is there. What is proven: control walked every 15 minutes through tomorrow with its own model of the packs, not the curve's levels (_mode_roles() each quarter; the house from the profile, the sun at 100 % trust, a GIVE role counted only where control's own discharge envelope packshare.envelope() is open, the cut-out at the floor), charges and gives in exactly the quarters the tab has a block in 13 cases (test_ux54.py part 7, own_walk()), e.g. 25/45 % with the peak guard: control gives 17:00-20:15, the block is 17:00-20:06; 30/80 % with a 7 kW evening house, B held to its rating: control 17:00-19:45, the block 17:00-19:32. The floor point is placed in one cut, where the battery runs out at the rate it gives (_floor_cut(), checked to the second against times worked out by hand), so a build stays small: at most 14 floor cuts and 28 level points per battery and day in those cases. With a cut-out under the floor, or a house or sun that differs from the forecast and profile, control differs (Limits).
• The sun is what solarfc reads. An hourly split (Solcast detailedForecast, Open-Meteo wh_period) is used as given. A daily total (Forecast.Solar, and any other state-only sensor, the owner's three "today remaining" and three "tomorrow") is spread over the 06:00-20:00 bell, and the tab says so.
• The house is HouseProfile.estimator(), never below the night median, as the plan reads it.
• The level did not exist (the plan holds targets and run blocks only), so daycurve.curve() derives it: each battery has the role modetrade.decide() gives (follow the house, hold, give, charge from the grid, rest), spare solar goes to Solar first batteries first, a shortfall is met by those that may give, shared as control shares it (by what each holds above the floor, packshare.weights(), each at most its rating, the rest passed on), both at dayplan.EFF, inside the plan's floor and ceiling, from the level now. Spare solar is trust x max(0, sun - house) and what is stored loses EFF, the plan's own trust x EFF x max(0, sun - house) (10 kWh of sun, 2 kWh of house, 70 %: 5.32 kWh stored). A shortfall is not scaled by trust. The tab says it counts on solar_trust_pct. Grid charging goes to modetrade.charge_to(). It is an estimate.
• Measured solar is kept per local hour in memory (SolarToday), from the reading the control cycle already takes; nothing is read back.
• On request. GET forecast builds the tab (Controller.forecast_payload() reads /status' plan first, so today's blocks are the very ones Home has). No Home Assistant call, nothing cached, nothing in the cycle but one in-memory add. The page asks every 30 s while the tab shows.
• Not known yet is a note, not an empty chart: modes off, no forecast set, forecast missing or old, learning, no tomorrow forecast, a monthly cycle, no level reading, control stopped, shadow.

What does not change

No decision, command, timing or option key. The one write in the control path is SolarToday.add() after the solar read, in a try, in memory.

Limits

• Prices are planned about a day ahead (priceruns.HORIZON), so a block of tomorrow later than 24 h from now, or one that needs prices not yet published, appears as the day comes near. The tab says so.
• Today's blocks are Home's, and are not extended or shortened: control may plan more once they end, and Home draws it then. The curve stops charging at its level (charge_to()) and giving at the floor.
• Today, a give run in the plan on a battery that is empty now has no block on Home (no energy to give). The tab equals Home, and says so: "The add-on plans to give power. No block shows while a battery has no energy to give." The curve gives in that run if the sun fills the battery first.
• With no forecast for tomorrow the curve stops at the first hour of the sun, solarfc.BELL_FROM_H (06:00): Forecast.between() has no answer past it. The note names that hour from the constant.
• Tomorrow shows price blocks only. A monthly cycle is drawn today, where the curve then stops for that battery.
• The curve uses the arbitrage floor and ceiling, as the plan does, not the battery's own cut-out level. Control stops a pack at its cut-out, so with a cut-out under the floor a battery in a give run covers the house below the floor; Home draws no block there either. The house profile is a median per hour, net of the solar the house used.
• Modes off: the count knows no sun and no house. A battery at the floor that the sun fills before a give gets no block until then (the tab says "The add-on plans to give power. No block shows while a battery has no energy to give."), and a small charge may be used by the house before the give. The other way round, a charge is drawn even when the sun fills the batteries before it: at the ceiling control then takes none (a cheap quarter at 13:00 on a sunny day: the tab draws the charge, control charges nothing at 95 % or 100 %). A leg is the verdict's as it stands now, one price period, as on Home; with several equal prices in a row control takes the next one as each passes.
• Measured solar starts when the add-on starts; the tab says from when.
• Clock changes: the tab walks local hours with dayplan._hour_slices(), which steps in UTC (SAFETY-123), and keys measured solar by the start of the local hour taken back from the UTC instant, so the spring day has 23 hours, the autumn day 25 (two of them 02:00), and a request always returns (test_ux54_dst.py, every quarter of both change days, with and without tzdata). The solar bell is spread over hours from local midnight, so on those two days it is an hour off the clock (SAFETY-123, not changed here).

Found, not changed

dayplan.plan() reads the forecast hour by hour with Forecast.between(s, e). For a state-only forecast between() takes its start as the moment the "left today" figure was read, so each later hour is spread over what is left of the bell from that hour: on the owner's shape 14.4 kWh left gives 57 kWh over the hours to midnight (test_ux54.py part 1). The plan's sun is overcounted for those sites. It does not show on a site whose batteries are all Solar first (no level is planned for them). The tab measures both ends from now (dayview.sun_kwh()), which keeps the total and gives the same hours for an hourly split. A card should fix the plan.