We took our emission factors from summaries of the Defra data, like most people do. Then we read the workbook itself. This is what we found, including in our own numbers.
Why we went looking
Every carbon figure a UK contractor reports rests on an emission factor: litres of red diesel multiplied by a number, tonnes of concrete multiplied by another. The numbers come from the DESNZ/Defra greenhouse gas conversion factors for company reporting, published annually.
Almost nobody reads the workbook. It is around forty sheets, and the same fuel appears in several of them at different units. So most people — us included — take the values from a summary: a blog post, a consultancy's cheat sheet, another tool's documentation.
We rebuilt our factor set from the workbook itself. Every factor that could be checked against it was close but not exact, and grid electricity was out by 35%.
This is not a story about someone else's software. These were our own numbers, taken the way most people take them. The useful part is why they were wrong, because the causes are structural and they will affect any tool sourced the same way.
What we found
1. Grid electricity, out by 35%
The largest single discrepancy, and the one that matters most because electricity is Scope 2 for every site with a cabin, and it is the factor that changes fastest as the grid decarbonises.
A summary that was accurate when written goes stale quietly. Defra republishes annually; the UK electricity factor has moved substantially year on year. Nothing in a copied number tells you which year it came from.
2. Red diesel priced as road diesel — about 8% under
Defra publishes red diesel as "Gas oil", a separate row from road diesel. Ours had been set to a road-diesel value of 2.527 against the correct gas oil figure of 2.75541 kgCO2e/litre.
About 8% under. That matters more than the percentage suggests, because red diesel is usually the largest single Scope 1 source on a construction site — plant, generators, welfare units. An 8% error on the biggest line is not a rounding difference.
3. Freight, roughly six times too low
The workbook gives HGV emissions per kilometre at several load factors — 0%, 50%, 100% and average laden — and also has an entirely separate sheet expressed per tonne-kilometre.
Pick the wrong column and you are not slightly out, you are out by a multiple. Ours had been roughly six times too low. The correct read for a delivery where you know the distance but not the payload is the average laden per-km column, converted to miles.
4. CNG: per kilogram or per litre?
The same fuel appears at 2.57546 per kg and 0.4507 per litre. Both are real Defra figures answering different questions. If your form collects kilograms and your factor is the per-litre row, you under-report by about six times.
This is the general trap: the workbook is not a list of fuels, it is a list of fuel-and-unit combinations, and a summary that drops the unit column produces a number that looks fine and is not.
5. Materials had no source at all
The worst finding, and the least visible.
Our material factors — concrete, steel, insulation, and the rest, in kgCO2e per tonne — carried a citation pointing at a benchmarks table. That table documents building-level benchmarks in kgCO2e per m²: a completely different quantity. The citation never covered those rows. It only looked as though it did.
An obvious gap gets found. A citation that is present but wrong does not, because nobody goes looking.
Defra cannot fully settle these either. Its Material use sheet is cradle-to-gate, which is the right quantity, but it is material-stream level. One "Insulation" figure of 1861.803 covers products spanning mineral wool around 1050 to PIR around 3500. "Metals" at 3821.949 is a mixed stream, not structural steel. It can check twelve of thirty-one rows and no more.
Where Defra does settle it, we now use it: timber (Wood, 269.50416) and glass (1402.76667), both identical across the 2025 and 2026 workbooks. The rest are marked as interim in our reports until a product-level dataset replaces them.
6. A hazardous waste factor that did not exist
There had been a hazardous waste factor of 50 kgCO2e per tonne in our code for years.
No Defra workbook publishes one. We searched all forty sheets of both the 2025 and 2026 releases. The number was invented — the last unsourced value in the waste model.
WRAP cannot supply it either. The 2026 waste sheet points users to WRAP's Carbon Waste and Resources Metric and then states outright that "outputs from this metric are not suitable for reporting Scope 3 emissions."
So we removed it. Hazardous tonnages are tracked and the consignment note is held, but no emissions figure is produced. No number is more honest than a number nobody can source.
Why secondary sources drift
Four causes, none of them carelessness:
Units. The workbook publishes many fuels by tonne, by litre and by kWh. A summary usually picks one and drops the column heading.
Load factors. Freight is published at several load factors and in two different denominators. Only one is right for your question.
Aggregation level. "Insulation" as a stream is not the insulation you specified. Stream-level figures are correct for their purpose and wrong for product-level accounting.
Time. Factors are republished annually. A copied value carries no year, and the ones that move most — electricity above all — are the ones people copy most.
None of this makes the summaries dishonest. It makes them unsuitable as the source of a number you put in a statutory return.
How to check the tool you use
Five questions. Any tool that cannot answer them is giving you numbers you cannot defend.
1. Which factor set year, and does it follow the activity date? Emissions should be calculated with the set current for the date of the activity, not the date of the report. If your tool applies the latest set to historical activity, your prior-year figures change every time Defra publishes — and a restated total is not a correction, it is a different answer to the same question. That is awkward when the original went into a SECR filing.
2. Which row, for red diesel? It should be Gas oil, not road diesel. Ask, and ask which unit.
3. What happens to hazardous waste? If it returns a tidy per-tonne figure, ask for the source. There isn't one.
4. Where do the material factors come from? "Industry averages" is not a source. Defra's Material use sheet is a legitimate answer for some streams, and a named product EPD is the best answer; a number with no provenance is not. Be careful with ICE specifically: Circular Ecology ends non-educational use of the ICE Database on 30 September 2026, so a tool still citing it after that date is either licensing its commercial successor or citing something it is not entitled to use.
5. Can it show the working? Quantity, factor, unit, source, year. If a report gives a total with no derivation, nobody can check it — including you, when a client asks.
What we changed
Factors are now transcribed from the DESNZ/Defra full workbook, read from the source sheets rather than a summary. Two sets are held — 2025 and 2026 — and the date of the activity selects the set, so a figure already submitted does not move when a new workbook lands.
Where a factor cannot be sourced properly, we produce none and say so. The three we still cannot settle are published in the open: material factors partly interim pending a product-level dataset, hazardous waste not carbon-weighted, and on-site recycled water, rainwater harvesting and tankered delivery marked as our own modelled estimates rather than government data, because Defra publishes no figure for them.
If you take one thing from this: ask where the number came from. Not which standard it claims to follow — where the number came from. It is a fair question, it takes a minute to answer, and the answer tells you most of what you need to know about the tool.