Understand the estimate.

Start with the device you actually use.

An activity category helps organize the inputs. It does not establish a device’s wattage. For a constant-power scenario, you enter operating watts, active duration and quantity. For a cycle, enter the declared energy for that programme and its complete duration.

A label, your own estimate or a measurement you report can supply the energy input. A reported measurement remains user-provided; CarbonShift does not connect to or read your equipment.

Exact references, with their conditions.

The current catalogue has 4 reviewed appliance references. Each retains an exact model, source, version, review date and operating conditions. Rated or maximum power describes a scenario, rather than measured average use.

Bosch WGE03408GB · Eco 40–60 label average

Weighted label value of 46 kWh per 100 Eco 40–60 cycles, divided by 100. This is a rounded programme average, not the energy of every load. Enter your actual programme duration; energy is assumed uniform during that duration. Inspect the source · reviewed 2026-10-23 · version 1.

Samsung UE50RU7455U · declared on-mode power

Supplier-declared on-mode consumption of 91 W under the listed television test conditions. Your settings and content can change consumption. This reference does not include standby time or identify other TV models. Inspect the source · reviewed 2026-10-23 · version 1.

Philips FC9744/09R1 · maximum-power scenario

Maximum input power 900 W, with electronic power control. This models active runtime at that rating; it is not measured average use and may overestimate a lower setting. Inspect the source · reviewed 2026-10-23 · version 1.

Kärcher K 2 Power Control · connected-load scenario

Connected load 1.4 kW. Model only active motor runtime at this rating, not the full time spent cleaning. Follow the equipment instructions for the surface. Water, detergent and other lifecycle effects are excluded. Inspect the source · reviewed 2026-10-23 · version 1.

You can use other electrical devices through custom input when you provide enough power or cycle-energy information. Unknown energy is a request for more information, never an automatically invented value.

A changing grid, with a clear source.

CarbonShift uses the NESO Carbon Intensity API for Great Britain’s national signal, attributed under CC BY 4.0. Forecast and actual-estimate fields describe half-hour intervals. These are national-average grid factors, not a reading from your home or a measurement of the marginal emissions caused by your appliance.

The simulator applies the factor for each part of the activity. Future scenarios use forecasts; completed scenarios use actual grid estimates where available. A task spanning several intervals uses all of its relevant overlaps, rather than only its start interval.

Source and retrieval time travel with the result. Current home-page context is cached for five minutes. Accepted plans keep the exact source snapshot reviewed at acceptance, rather than silently acquiring a newer forecast.

Keep small activities precise.

The household method stores integer milliwatts, milliwatt-hours and milligrams of CO₂. It integrates energy segments over their exact overlap with the grid’s time intervals using rational arithmetic.

Declared cycle energy is spread evenly across the entered duration unless a reviewed supported source supplies a different profile. Changing a cycle’s duration while keeping its kWh fixed preserves total energy. Changing watts or duration changes the energy in a constant-power scenario.

AI can suggest editable fields and select grounded explanation codes. Deterministic software validates constraints, integrates energy, compares schedules and verifies proofs. A model cannot assign authoritative watts, overwrite totals or accept a plan on your behalf.

Compare a routine you could follow.

The baseline is the explicit time you would otherwise use for every task. You review it before comparing. Each activity must fit completely inside an allowed window, avoid its blocked periods and wait for its predecessors. Tasks sharing equipment or requiring your attention cannot overlap. Fixed appointments remain fixed.

Duration, energy and quantity stay constant across timing alternatives. The difference is baseline estimated grid CO₂ minus the selected estimate. Zero improvement is a valid result; a negative updated difference is a higher estimate. Percentages are unavailable when the baseline is zero.

Starts that are actually evaluated

The engine considers 5-minute UTC starts, legal window and block boundaries, the explicit baseline and fixed starts. It does not claim to search every possible instant.

When tasks interact

Coupled search is bounded at 50,000 candidate extensions. A completed search is optimal within the evaluated candidates. A limited search shows “best found”; a better feasible schedule may exist.

Alternatives and longer windows

Up to 3 distinct alternatives come from evaluated feasible schedules. A separate window-extension proposal preserves the energy and baseline, uses one fresh forecast for that proposal, and needs a new review before saving.

Full planning requires forecast coverage throughout the searched range, including blocked periods, within the available 48-hour horizon. An insufficient window, conflicting baseline, missing source and exhausted search have separate outcomes.

A London day means a London day.

Scheduling uses Europe/London regardless of your browser’s location. Records keep explicit UTC instants; screens show the London date, clock time and offset. When the clocks repeat an hour, you must choose which occurrence you mean. A nonexistent local time is rejected.

A calendar day runs from local midnight to the next local midnight. That can be 23, 24 or 25 elapsed hours. A day plan keeps every task and availability window within those boundaries; it never extends into the next day silently.

Comparisons expire after at most 5 minutes and no later than 10 minutes before the earliest relevant task start. The server rechecks time and source freshness at acceptance. Simulation can model an immediate activity; saving a future plan needs room to review it.

Reports add context without rewriting history.

Saved inputs and forecasts remain immutable. Report each task separately after its full activity finishes, or report that it will not run. A correction appends a new report; the latest eligible result for each task is counted once.

An updated comparison uses actual grid factors for both the original baseline and the reported time, with the same accepted energy assumptions. If duration or device energy changed, the report stays in history without a comparable updated difference. A non-run earns no avoided-emissions amount.

Execution remains self-reported. Actual grid factors improve the calculation’s source basis; they do not establish that a device ran, measured its consumption or achieved a physical emissions reduction.

A receipt keeps the reasoning inspectable.

Private saving and public publication are separate decisions. Exact consent discloses the reviewed public manifest: sanitized task identities, timing, availability, energy assumptions, source inputs and calculation details. Timing can still reveal a routine.

The Sepolia contract records versioned digests and revision order. A live plan commitment must precede the earliest baseline or selected start. Each public result revision needs separate consent. Private corrections never publish automatically.

The verifier checks format, recorded commitment, known method, calculation, timing, result lineage and evidence disclosures separately. Without trusted chain observations, chain-dependent checks remain unavailable. A matching digest binds content; it does not certify the truth of inputs or measured avoided emissions.

Existing records keep their method.

Earlier washing plans use carbonshift-gb-uniform-v1 with their original Wh profiles. Household schedules use carbonshift-gb-activities-v2 with mWh. Explicit version dispatch preserves each record’s units, calculation rules and hash prefixes.

An unknown method is unsupported rather than guessed. Opening an older record never upgrades its calculation or changes its accepted assumptions silently.

What this model leaves outside.

CarbonShift estimates direct electricity and grid-associated CO₂ for reviewed scenarios. Water, fuel, detergent, product manufacture, tariffs, payments, carbon credits, appliance control and meter verification are outside this calculation.

The current carbon signal is national Great Britain data. It is not local guidance for another country, regional routing, Northern Ireland or a guarantee of climate impact. Always use equipment in safe operating conditions and follow its instructions.