# Watts Where - one home's electricity, room by room, and a plan to cut this winter's bill

> Household electricity map. List your rooms and what plugs in under each one - "Space heater 1500W 5h", "TV 4h, standby 1W", "Dryer 4 loads a week", "Fridge from 2003" - with your price per kWh and, optionally, a time-of-use rate and last month's kWh from the bill. The page prices every device (running plus standby) and draws the home as a floor plan where each room is as big as what it uses. A metered plan gives each room a call and ranks the moves by the kWh they save.

URL: https://watts-where.skillsafe.ai/
API: https://watts-where.skillsafe.ai/api.html

Built on the SkillSafe skill @personamanagmentlayer/energy-expert (energy systems, power management, demand response and smart-meter analytics), brought down to one household.

## What runs free in the browser (wattkit.js, housemap.js)

- Reading each line: watts or kW, quantity ("2 space heaters", "x3"), hours a day, minutes, uses a week ("4 loads a week"), miles a day for an EV, standby watts, a year for fridges and freezers, and when it runs (overnight, after dinner, at 7pm) against the time-of-use peak window.
- Estimates, always labelled: wattage, hours, duty cycle and standby draw of any device whose line does not state them come from a table of typical values; the price of a kWh is 0.18 when left blank; an EV uses 0.30 kWh a mile; a month is 30.4 days.
- kWh and cost per device, per room and for the home; standby kWh separately; the share of each room and category.
- Flags: heating 30% or more of the list, electric heat with no heater listed, standby of 15 kWh a month or more, old or second fridges and freezers, incandescent and halogen bulbs, a large water heater, hot tub or dryer, shiftable loads at peak (with a time-of-use rate), devices with no wattage, and a gap of more than 25% against the bill.
- The floor-plan map (squarified rooms sized by kWh, coloured by share, standby hatched), a ranked device list, and exports: Markdown, a devices CSV and this week's steps.

## The metered task (model gpt-terra)

- `plan` - inputs `today`, `currency`, `home`, `rooms`, `devices`, `page_reads`, `page_verdict`, optional `notes` (object fields JSON-encoded as strings). Returns `title`, `verdict` (lean|trim|overhaul), `summary`, `rooms` [{room, action keep|trim|shift|swap, why}], `moves` [{device, kind trim|shift|swap|unplug, move, kwh_month, effort free|cheap|invest}], `bill_check`, `this_week`, `caveats`.
- The page's check: every move must name a listed device; a saving is capped at what the device uses (all moves on one device together), an unplug at its standby kWh, a shift at its running kWh; shifts are priced only with a time-of-use rate; rooms with no call are shown as keep; a lean verdict over a large lever is flagged. The page prices every move itself at the person's rate - the model never writes money.
- House rules: only the input's numbers; no brands, prices, shops or rebate amounts; no advice that trades warmth or fire safety for savings.

## Limits

- Estimates, not measurements. A plug-in meter or the utility's usage data is the check. Heating with gas, oil or propane is not on this electricity list.
- Context: the US Energy Information Administration's Winter Fuels Outlook (October 2026) expects electric-heat households to spend about 4% more this winter (https://www.eia.gov/outlooks/steo/report/perspectives/2026/10-winterfuels/article.php). Sample homes are made up.
- Saved homes go to the signed-in user's own collection (`homes`), only when they press save.