Commonly handled layouts include:
- Separate date and time columns
- Combined timestamps
- Half-hourly readings
- Grid format (dates down the rows, intervals across the columns)
- Settlement periods such as P1–P48 or 1–48
- Other column orders and naming styles
You can also tag the site with an industry: School, Agriculture, Manufacturing, Sports facility, Warehouse, or None.
Learn more about industry tagging
- Identifies the date, time and consumption fields
- Converts everything into one standard format
- Adds half-hourly readings up into hourly consumption
- Handles UK daylight-saving clock changes
- Flags missing readings
- Produces a consistent 365-day profile
Look over the consumption graphs and the validation summary before moving on. If anything about the detected structure looks wrong, open the detected-columns details on the check screen and confirm the fields before continuing.
- Actual data — measured readings from your file, including genuine 0 kWh readings
- Missing — hours with no reading at all (blank or absent), left empty rather than zero
- Extrapolated — modelled hours, kept in their own column
A 0 in your file is treated as a real reading of zero consumption, so it is never overwritten by extrapolation. Only hours with no data are counted as missing.
The missing-periods table lists the dates involved. If the dataset is complete you can go straight to export; otherwise choose an extrapolation method.
Only fills hours where data is missing. Existing actual readings are never changed, and months whose actuals already exceed the estimate are left alone.
Use when part of the year doesn't represent normal site operation. Mark a reference period that does; the pattern learned there is used to top up the rest of the year.
Ranges use day and month only, and a range that ends before it starts (December → February) wraps across the year. Actual readings always stay separate from generated values.
Learn more about the extrapolation chat
The library collects approved datasets per industry as normalised shapes: hourly patterns, weekday versus weekend differences, daily patterns and seasonal patterns. Absolute consumption from another site is never reused.
Learn more about how the library is used today
- Actual consumption — purple in the charts
- Extrapolated consumption — green in the charts
- Final modelled consumption — the two added together
The coverage panel shows the percentage of the year that is measured versus extrapolated, and any reference period you marked is listed above the charts. If the result doesn't look right, use Reset and try a different method.
Two columns — DateTime and Energy (kWh), where energy is the final total for that hour.
Four columns — DateTime, Energy (kWh), Extrapolated (kWh) and Total (kWh), so measured and generated values stay distinguishable.
Exports are named after the project reference entered at upload, e.g. PRJ0123 Solar Edge.CSV.
The Model quality panel below every generated profile shows:
- Reconciliation — annual, monthly, day/night, daily and hourly totals must each match the level above exactly.
- Hard constraints (your totals, night window, measured readings) are never altered; soft constraints (seasonal, weekday and hourly shapes) only decide timing.
- Shape contribution — how much of the shape came from the site's own measured data versus the industry library, based on measured coverage.
- Confidence — an arithmetic score from coverage, known constraints, contributing reference sites and library variability. It is never AI-generated.
A failed structural or reconciliation check blocks export until the inputs are fixed. Use Model accuracy in the sidebar to hide a known period from an approved dataset, rebuild it with the live model, and review energy, profile, peak and day/night error per industry, plus regressions between calculation versions.
Switch to Manual on the upload screen, then:
- Enter the project reference, dataset name and industry as usual.
- Choose Annual total (one figure, in kWh, MWh or GWh) or Monthly totals (a 12-month grid). Blank months are estimated from the industry shape and flagged as Estimated in the reconciliation table.
- Optionally add a Day / Night split — enter the night energy or a night percentage and set the night window (default 00:00–07:00). The generated profile matches both the day and night totals exactly.
- Press Generate Consumption Profile. You get the same 8,760-row 2025 calendar, graphs and exports as an uploaded file.
Distribution across months, days of the week and hours comes from the industry reference shape; the magnitude comes only from the totals you entered. All arithmetic is deterministic, and every export row carries the input method, monthly target, target source and shape version so the result is auditable.
My dataset is complete
Review the graphs and export.
A few hours or days are missing
Use Fill Missing.
Several weeks or months are missing
Use Fill Missing, and note in the review that a large share of the year is modelled.
Part of the year is reduced or abnormal operation
Use Add Shortfall and mark a normal operating period as the reference.
I have very little representative data
Extrapolate if you must, but treat the result as low confidence — the coverage percentages show how much is generated.
Completed datasets can be approved for the Reference Library, contributing their anonymised consumption shape to the selected industry.
- Only approved, actual measured data contributes.
- Extrapolated or generated values are never used to build reference shapes.
- Another site's absolute consumption values are never used when extrapolating a new site.