How we calculate
your carbon emissions.
Every number on every report traces to a published factor, stamped with the engine version that produced it. No black boxes. This page documents the methodology end-to-end, from the CSV you upload to the final emissions figure in your PDF, with every step disclosed.
The GHG Protocol Corporate Standard.
Every Carbon Draft report aligns to the GHG Protocol Corporate Accounting and Reporting Standard (Revised Edition, 2004) and its Scope 2 Guidance (2015). In 2016, 92% of Fortune 500 companies responding to CDP used the GHG Protocol directly or through a program built on it. Developed by the World Resources Institute (WRI) and the World Business Council for Sustainable Development (WBCSD), it underpins CDP, the Science Based Targets initiative (SBTi), California SB-253 and the EU CSRD. When a supplier questionnaire or disclosure filing asks for Scope 1, 2, and 3, this is the accounting grammar they mean.
For Scope 3 we also apply the Corporate Value Chain (Scope 3) Standard (2011) and its Technical Guidance for Calculating Scope 3 Emissions (2013), which sets out the spend-based method: the money spent in a category multiplied by an environmentally extended input-output emission factor.
Scope 1 · Direct emissions
On-site fuel combustion (natural gas, diesel, gasoline), company-owned or controlled vehicles, process emissions from owned equipment, fugitive refrigerant leaks from HVAC. Any GHG released directly by equipment you own or operate.
Scope 2 · Purchased energy
Emissions from the generation of purchased electricity, steam, heating or cooling. We apply location-based factors: in the US, EPA eGRID 2023 by state (by subregion, such as CAMX or RFCE, when you give kWh); elsewhere, the country's own published factor, by state or grid in Australia and Japan. Per GHG Protocol Scope 2 Guidance (2015). When you give your total kWh with certificates (RECs, GOs, PPAs) or a supplier's rate, the report adds a market-based figure.
Scope 3 · Value chain
15 upstream and downstream categories. From a ledger the engine prices the ones your spend shows: 1 (purchased goods and services, including cloud and software), 2 (capital goods), 3 (grid losses, derived from your electricity), 4 (upstream freight), 5 (waste), 6 (business travel) and 7 (employee commuting), plus 12 (end-of-life of sold products) when you give tonnages. Complete and Disclosure-Aligned reports carry the same Scope 3.
A dollar becomes a kilogram.
Spend-based carbon accounting does something simple by design: every dollar spent in a commodity category is multiplied by an industry-average emissions intensity (kg CO₂e per dollar) to estimate the embedded emissions. EPA's Supply Chain GHG Emission Factors v1.3, built on its USEEIO model, supply these intensities for 1,016 commodities by six-digit NAICS code; the engine maps your purchases to about 70 spend categories, each tied to one or more of those codes.
Outside the US the same dollar carries a different footprint, because production, grids and prices differ. A UK company's Scope 3 uses the Defra / University of Leeds spend-based multipliers (kg CO₂e per pound, by SIC product group). Every other country uses CEDA by Watershed: the Open CEDA input-output factors for the company's own country (148 countries, 400 sectors), which we convert from producer prices of CEDA's base year to purchaser prices of the year of the spend. Each report prints the licence notice, and each ledger row names the CEDA release and country used.
| kg CO₂e per USD | US | IE | DE | NL | SE | IN |
|---|---|---|---|---|---|---|
| Management consulting | 0.081 | 0.081 | 0.083 | 0.075 | 0.056 | 0.620 |
| Computers and electronics | 0.096 | 0.234 | 0.228 | 0.215 | 0.190 | 1.231 |
| Road freight | 0.595 | 0.184 | 0.261 | 0.297 | 0.340 | 0.946 |
This trades precision for coverage. An activity-based measurement (therms of natural gas burned, gallons of diesel pumped, kWh of electricity metered) is accurate to roughly ±5-10%; a spend-based estimate of the same category, roughly ±25-40%. Both ranges are our engineering estimates; the GHG Protocol publishes neither. The tradeoff: activity data is expensive and slow to collect; spend data already sits in your general ledger and can be classified in minutes. For a screening-grade response to a supplier questionnaire or an internal baseline, spend-based is the correct tool. California SB-253 lets reporting companies use spend-based and other secondary data for Scope 3, and the draft rules CARB previewed at its July 2026 workshop do not require Scope 3 assurance, so a spend-based report can feed a customer's SB-253 Scope 3 disclosure. Where figures will be assured by a third party, high-impact categories should move to activity data.
Credits, refunds and returns reduce the spend of their own category before the factor is applied, so each category is priced on what you actually paid. A category whose refunds exceed its purchases is set to zero, not below. The data quality section counts the credit rows, and the Disclosure-Aligned report lists each one, marked CREDIT, in its ledger.
The factor sources.
We combine published factor sets, and every report's cover names its factor pack and the sets in it. The main ones are below. Electricity factors outside the US come from a national publication where we hold one (for example DESNZ for the UK, SEAI for Ireland, NGA for Australia and CEA for India), from the EEA for several EU grids, and from the IEA World Energy Outlook 2024 for the rest. Fuel spend priced at a local fuel price uses the national list where we hold one (DESNZ, SEAI, CO2emissiefactoren, NGA) and the EPA Hub otherwise; fuel quantities entered directly use the same national factor where we hold one for that fuel, and the EPA Hub otherwise. No proprietary multipliers, no black-box coefficients. Every rate below is reproducible from public data, and the values for each supported country are listed on the emission factors pages.
Every citation badge throughout this page (and the rest of the site) opens the methodology drawer on the right, showing the factor's exact rate, source publication, region, the engine version that applies it, and a plain-language explanation. Click any natural gas, electricity, or cloud services cite to try it.
Vendor to commodity code.
Every vendor in your CSV needs to be mapped to a NAICS commodity code before a factor can be applied. The keyword classifier runs these layers in order, each catching what the previous one missed. The shares are its run, on its own and without Claude, on the 266-row ledger behind our published US sample, the fictional Scioto Valley Foods LLC:
- Regex pre-pass: 39 compiled patterns for accounting-system transaction prefixes (
AMZ*,SQ *,PAYPAL *,SHELL-OIL,MSFT AZURE, airline ticket prefixes, ride-share tags). Catches 10.5% of rows. - Normalized substring: vendor name is stripped of LLC/INC/LTD suffixes, transaction refs, long digit runs, then matched against a 393-entry known-vendor dictionary. Catches 20.3%.
- Fuzzy match: rapidfuzz partial ratio with a threshold of 85. Catches typos (
chevorn→ Chevron) and truncations: 0.0% of this ledger, which has few. - Account and description keywords: the account name and the line's description, read for what was bought. Catches 40.2%.
On every Carbon Draft, each row in the reporting period is classified by Claude, for what the keyword layers cannot read: vendor normalization ("SHELL OIL STN" → Shell), ambiguous categories and non-English supplier names. Vendors you have already identified for us use your answer. The keyword layers above classify any row Claude leaves open, the rows past the first 2,500 and rows outside the reporting period, and every row when the AI service is unavailable. Each row gets a confidence score (0.0-1.0); the Data Quality Summary shows the bracket each row landed in, and the methodology states how many rows Claude classified.
On the same ledger the keyword classifier gives 48.5% of rows high confidence (≥0.80) and 22.6% medium; 28.6% are payroll, tax, rent and the like, excluded as not emissions, and the other 0.4% it cannot recognise. Those rows are listed in the report as unrecognized and not priced, for the customer or an auditor to resolve, never guessed or silently hidden.
What the intake layer handles.
The upload accepts CSV and Excel (.xlsx, .xls) exports from any general ledger, native SIE 4, XAF, DATEV, Yuki and Tally exports, and bank and card statements. A CSV or Excel file needs two columns (vendor/description + amount); everything else is optional. The intake layer auto-detects:
- File encoding: UTF-8 and Windows-1252 (the latter is a common QuickBooks export default that breaks most parsers)
- Decimal convention: US (
1,234.56) vs European (1.234,56) - Negative notation: hyphen (
-1234) vs parens-accountant ((1234)) - Localized headers: German (
Lieferant→ vendor,Betrag→ amount), French, Spanish, Dutch, Swedish and Belgian French header maps - Date formats: MM/DD/YYYY, DD/MM/YYYY, YYYY-MM-DD ISO, free-form strings
- Currency: amounts are shown in the ledger's own currency, with US dollars beside them. Spend factors per US dollar take the dollar amount at the static rate stamped on the report (current: February 2026 rates); the Defra multipliers are per pound, Open CEDA converts at the ECB rate for the month, and fuel and mileage at official prices in the ledger's currency are read from the amount as booked
Every decision the intake layer makes is logged to the Data Quality Summary so auditors can reconcile. If the parser has to guess (e.g., a file with no column headers), the guess is disclosed.
What we don't know.
Spend-based accounting is a screening method, and the GHG Protocol accepts it: the Scope 3 Standard lists money spent as activity data (Table 7.2), and its 2013 Calculation Guidance sets out the spend-based method. The uncertainty figure attached to it here, roughly ±25-40% at the category level, is our engineering estimate for the method; the GHG Protocol publishes no such figure. Activity-based data (therms, gallons, kWh) is tighter, on the order of ±5-10%.
- Use EPA's Supply Chain GHG Emission Factors v1.3 (published July 2024) and EPA eGRID 2023 (rev2, June 2025), and stamp the versions on every report
- Net credits, refunds, returns and billing adjustments within their category before the factor is applied
- Disclose per-row confidence in every report's ledger: the top 20 rows by spend on Basic and Complete, every priced row on the Disclosure-Aligned Report
- Stamp the factor pack on every page of every PDF, and the engine version that produced the numbers on the cover, so prior reports remain reconcilable on their original methodology
- Support activity-data override (gallons, therms, kWh, refrigerant kg) for tighter accuracy where the customer has the data
- Log every intake decision (encoding, decimal, header-map) in the Data Quality Summary
- Apply proprietary multipliers or undisclosed adjustments
- Pretend spend-based estimates are activity-based precision
- Substitute for an external audit or assurance engagement
- Invent factors no published source supports
- Silently drop rows we can't classify. Low-confidence rows are surfaced for manual review
When a Carbon Draft is enough.
A screening-grade Carbon Draft is the right tool for:
- Responding to supplier sustainability questionnaires (Walmart Project Gigaton, Microsoft Supplier Sustainability, Apple Clean Energy, CDP Supply Chain)
- Establishing a carbon baseline for internal strategy
- Identifying emission hotspots before making reduction investments
- Preparing for a third-party assurance engagement (the Disclosure-Aligned Report tier front-loads much of the preparation work an auditor would bill for)
It is not a substitute for:
- Third-party assurance (ISO 14064-3) on regulated filings
- Science-Based Targets initiative (SBTi) validation
- Activity-data-grade precision where a regulator or assurer requires it (for example a large company's own CSRD disclosures)
The engine history.
Every change to the classifier, factor tables, or intake layer is stamped and versioned. Prior reports remain reconcilable on their original engine version. We don't silently rewrite history.
Energy prices v1.3. A fuel quantity the customer supplies (litres of petrol or diesel, kWh of gas) is converted with the country's own factor wherever one is held for that fuel, the factor the same fuel bought by spend at a local price already used: DESNZ 2026 for the UK (diesel 2.58354 and petrol 2.075 kg CO2e per litre, average biofuel blend; natural gas 0.18231 per kWh, gross calorific value), SEAI for Ireland (CO2 only, as SEAI publishes no methane or nitrous oxide factors; an Irish diesel bill moves from 2.753 to 2.682 kg per litre), CO2emissiefactoren for the Netherlands and NGA for Australia, and EPA's figure elsewhere. A row on SEAI's factor counts wholly as CO2 in the split of Scope 1 by gas. UK heating oil bought by spend is converted into litres at DESNZ's price for standard grade burning oil (Quarterly Energy Prices table 4.1.2, excluding VAT) and into emissions at DESNZ's burning oil factor, 2.54016 kg CO2e per litre, in place of EPA's factor per dollar at a US price. LPG bought by spend stays on EPA's factor per dollar, as no official UK price series is published, and a UK Carbon Reduction Plan says so.
Factor Pack v2.6.17. Refrigerants use IPCC AR5 100-year GWPs, the set the engine already used for methane and nitrous oxide from combustion and that the EPA spend factors are built on (Open CEDA 2025, used for Scope 3 outside the US and UK, is on AR6): R-410A 1,924 (was 2,088 from AR4), R-134a 1,300 (was 1,430), R-32 677, R-404A 3,943 and R-454B 467, as tabulated in or derived from the EPA GHG Emission Factors Hub 2025, Tables 11 and 12, and R-22 1,760 from IPCC AR5 Table 8.A.1. Indian electricity uses the CEA CO2 Baseline Database v22.0 (August 2026): FY 2025-26, 0.6751 kg CO2 per kWh (was 0.7097 for FY 2024-25 from v21.0), and the rupee spend factor moves with it, from 5.93 to 5.64 kg CO2 per US dollar.
Factor Pack v2.6.16. Wood products have their own type: sawn timber, plywood, MDF and OSB sheets, chipboard, veneer and millwork at 0.192 kg CO2e per USD, the mean of the 14 wood-product industries (NAICS 321) in the EPA's Supply Chain GHG Emission Factors v1.3, with margins (sawmills 0.264, veneer and panels 0.208, millwork 0.154, other wood products 0.164). These purchases used to go unidentified or, for millwork, be read as furniture (0.212). A UK company's wood is priced by the Defra / Leeds multiplier for SIC 16, and a company elsewhere outside the US by Open CEDA's four wood-product sectors for its country. A row from a supplier the engine does not already know, whose description names timber or sheet materials ("Oak boards and plywood"), is read as wood, as are the German, Dutch, Swedish and French trade words (Schnittholz, Spanplatten, Eiken planken, Byggvirke, Contreplaqué). Wood yields to anything else a name or description says ("Timber Lodge Hotel" is a hotel, "Timber haulage" is road freight), and firewood, pellets and charcoal are not wood products. Haulage and haulier are read as road freight. The AI classifier is also told that a mileage claim for an employee's own car is business travel, not company fuel.
Factor Pack v2.6.15. A travel account that names no mode (Travelling Expenses, Travel - National, Staff Travel) is now priced instead of left out: at the mix of the air, ground and lodging travel the same ledger identifies, by spend, each part with the factor that mode gets for the company's country (Defra / Leeds in the UK, Open CEDA elsewhere outside the US, USEEIO in the US). A ledger that identifies no travel by mode takes equal shares, and the row says so. These rows stay listed so you can split them. Accounts that hold meals as well (Travel and Subsistence, Travel and Entertainment) and vehicle accounts (Car and Truck) are unchanged. Old-format Excel (.xls) bank statements are now read directly. Separately (mileage rates 1.2), Australian mileage from 1 July 2026 is read at the ATO's 91 cents per kilometre.
Factor Pack v2.6.14. Belgian electricity is the European Environment Agency's figure for Belgium in 2024: 155 g CO2e per kWh (emissions of public electricity and heat generation, heat removed, per kWh of gross production; network losses of 3.0%, Eurostat's, are counted separately in Category 3). It replaces 163 g, whose source could not be traced; no Belgian authority publishes an electricity factor for company reporting. A Belgian row with no dated price falls back to 0.613 kg CO2e per USD (the grid over Eurostat's latest non-household price, band IB).
Engine 2.5. Spend priced with the EPA's USEEIO factors, which are published per 2022 dollar, is moved to the dollars of the year of each purchase with the BEA GDP implicit price deflator (NIPA Table 1.1.9), as EPA advises: a 2024 dollar carries 6% less than a 2022 dollar and a 2025 dollar 8% less. This applies to a US company's Scope 3 and to any Scope 3 row no national provider covers; rows priced with CEDA or the Defra / Leeds multipliers were already in the prices of their own year. On our US test ledgers Scope 3 falls 5 to 7%.
Factor Pack v2.6.13. A purchase from a shop or a wholesaler is priced as the goods bought, as EPA advises, not as the shop: EPA's factors with margins already carry the retail, wholesale and transport margins. Office supplies move from the stationery-store factor (0.111 kg CO2e per USD) to stationery and office-supply manufacturing (0.281), groceries from the grocer (0.186) to food manufacturing (0.522), and general merchandise (0.164) and goods bought wholesale (0.120) to EPA's all other miscellaneous manufacturing (0.182). A marketplace or general retailer (Amazon, Walmart, Costco, Target) is priced as general goods (0.182), and an office supplier (Staples, Office Depot, Uline) as office supplies. Software subscriptions move from data hosting (0.093) to software publishers (0.080). CEDA factors outside the US and the UK already follow the same mapping. EPA's factors are also described as what they are: 2022 purchaser prices with margins, not producer prices.
Factor Pack v2.6.12. Scope 3 for a company outside the US and the UK is now priced with factors for its own country's economy: CEDA by Watershed (Open CEDA 2025, or Open CEDA 2024 for a reporting year ending in 2024 or earlier; CC BY-SA 4.0), in place of US factors. Each category takes the CEDA sectors of what was bought, matched as its US factor is, converted to purchaser prices of the year of the spend with CEDA's sector price index and, for goods, CEDA's purchaser-producer ratios plus the same country's trade and transport margins. Spend is converted to US dollars at the ECB reference rate for the month of each purchase. A country value that is zero, or more than three times or less than a third of that country's usual level against its region, is replaced by CEDA's regional average scaled to that level and named on the row. Scope 3 moves materially: on our test ledgers a German manufacturer rises from 87 to 155 t CO2e, an Indian services firm from 26 to 109 t and a Japanese company falls from 210 to 127 t. The report's note that Scope 3 is priced with US factors now appears only while rows remain on them.
Factor Pack v2.6.11. Transmission and distribution losses (Category 3) for 27 European countries are now Eurostat's own figures for 2024 in place of the US 4.2%: distribution losses per kWh available for final consumption, for example Germany 3.8%, Belgium 3.0%, France 6.5%, Spain 7.8%, Italy 4.6%, Poland 4.5%, Norway 4.0%. Ireland moves to the same basis, 6.9% (it used losses per kWh produced, 7.8%), which understates losses for a country that exports and overstates them for one that imports. Eurostat's 2025 figures, higher for most countries at once, are not used until confirmed.
Factor Pack v2.6.10. Swedish electricity is the Nordic mix, as Swedish guidance recommends for location-based Scope 2 (Fastighetsägarna, 2026): 46 g CO2e per kWh, the Scope 2 part of IVL Svenska Miljöinstitutet's Nordic mix for 2021-2023 with imports and exports (2025); the full life-cycle figure Naturvårdsverket uses is 59 g. It replaces 37 g, a Swedish-only mix. Network losses are Eurostat's for Sweden in 2024, 4.24% of the electricity delivered (Category 3), in place of the US 4.2%; the 2025 provisional figure, nearly twice every year since 2022, is not used until confirmed. A Swedish row with no dated price falls back to 0.345 kg CO2e per USD (the grid over Eurostat's latest non-household price).
India's petrol and diesel are priced where and when they were bought: each purchase becomes litres at PPAC's retail selling price for the month (the Ministry of Petroleum's daily table, mean of Delhi, Mumbai, Chennai and Kolkata; fuel is outside GST, so nothing is netted off), then emissions at the per-litre factor. Until now Indian fuel was divided by the US pump-price anchors, so the change follows the month's price: since 2022, when Indian petrol has cost Rs 95-110 a litre, Scope 1 on petrol falls by a quarter to a third and on diesel by 3-12%; in 2019-20, when diesel was cheaper, diesel rises. A purchase that names no fuel is read as diesel, 68% of India's petrol and diesel (PPAC consumption, 2026-27), where it was read as petrol.
Factor Pack v2.6.9. Australia on the National Greenhouse Accounts Factors 2026 (DCCEEW): the grid at 0.59 kg CO2e/kWh nationally and by state or grid when given (Victoria 0.74, Queensland 0.65, New South Wales and ACT 0.60, Northern Territory 0.55, Western Australia 0.45 and 0.56, Tasmania 0.23, South Australia 0.21), replacing the IEA figure of 0.681; network losses from NGA Table 40, the state's own when given and otherwise the mean of the eight grids (an estimate, as NGA publishes no national figure), replacing the US 4.2%; petrol and diesel per litre from NGA Table 9. Australian suppliers are recognised (Ampol, United Petroleum, fuel cards, Jetstar, Virgin Australia, StarTrack, Sendle, the water corporations, Cleanaway), dual-fuel energy retailers (Origin, EnergyAustralia, Alinta, Red Energy) follow the account and description, and superannuation, PAYG, BAS, fringe benefits tax, payroll tax, council rates and Linkt tolls are left out.
Factor Pack v2.6.8. Dutch accounts and suppliers are recognised when the AI classifier is not used: Elektriciteit, Stroom, Aardgas, Stadsverwarming, Brandstof and Tankpas, Verzendkosten, Afvalverwerking, Treinkaartjes, Telefoon, Drukwerk and the other standard Exact and Twinfield accounts; NS, PostNL, KPN, Renewi, Vitens and the water companies. Energy suppliers that sell both (Eneco, Vattenfall, Essent, Greenchoice) follow the account and description, as the Irish ones do. Payroll, BTW, corporate tax, depreciation, interest, bank charges, insurance and rent are left out. Swedish and French accounts and suppliers are recognised the same way (El för belysning, Drivmedel, Porto, Frakter; Électricité, Carburant, Mazout, Affranchissement, SNCB, Proximus), and a utilities account may be written in Dutch, French or Swedish ("Électricité et gaz").
Factor Pack v2.6.7. Dutch reports now use the Dutch national list, CO2emissiefactoren.nl 2026 (Rijkswaterstaat and Stimular): the grid mix at 0.200 kg CO2e per kWh including transmission and distribution losses of 4.06%, reported as 0.1922 in Scope 2 plus the losses in Category 3; petrol E10 2.139 and diesel B7 2.462 kg CO2e per litre, the blends sold at the pump; natural gas 1.779 kg CO2e per cubic metre (0.1821 per kWh on the gross calorific basis Eurostat prices gas in). The grid factor had been 0.328, a well-to-wheel figure that also counts the upstream emissions of the fuels burned, which overstated Dutch Scope 2 by 71%. When a row has no dated price, electricity now falls back to the grid factor over Eurostat's latest price (0.73 kg CO2e per USD in the Netherlands, 0.57 in Ireland, both previously about twice that). Dutch, Swedish and Belgian French column names are now read (Relatie, Omschrijving grootboek, Bedrag, Leverantör, Belopp, Libellé, Tiers and others): a Dutch export used to arrive with every supplier blank.
Factor Pack v2.6.6. A utilities account, one made only of utility words such as Xero's default "Light, Power, Heating" in the UK and Ireland or "Gas & Electricity", now holds every utility bill: the row's own description, then the supplier, says whether it is electricity, gas, heating oil, LPG or water. It used to name only "power", so gas, kerosene and water bills booked there were priced as electricity. An account that names one utility among other words ("Expenses / Utilities / Natural Gas") still decides, and a description or supplier naming two utilities leaves the account's own type. Suppliers of several utilities (Bord Gais, Electric Ireland, SSE Airtricity, Energia) never outvote what the row says; their main supply applies only when nothing else does. Irish suppliers, carriers and taxes are recognised: kerosene and home heating oil, Uisce Eireann, Aer Lingus, Irish Rail, Bus Eireann, Luas, An Post, Tax Saver commuter tickets (employee commuting) and Revenue payments (excluded).
Factor Pack v2.6.5. Ireland's factors now come from SEAI, the national energy authority: the grid at 178.3 g CO2 per kWh (2025 provisional: CO2 from generation in Ireland over gross electricity supply; grid losses are added separately at 7.8%), petrol 2.311 and diesel or gasoil 2.682 kg CO2 per litre, and natural gas 183.5 g CO2 per kWh on the gross calorific basis Eurostat prices gas in. The grid factor had been an international figure of 0.279 kg per kWh, which overstated Irish Scope 2 by 56%. SEAI publishes CO2 factors only, so methane and nitrous oxide from combustion are not included for Ireland, and each row says CO2 only. Reports now name the national source of the grid factor for Ireland (SEAI), the UK (DESNZ), Germany (UBA) and India (CEA).
Factor Pack v2.6.4. How a ledger row is read when the AI classifier is unavailable is now a fixed order: what you told us about a vendor; then an account that names a non-emitting use (a loan, bank charges, salary, tax); then an account that names one kind of purchase as its whole name or one part of it ("Strom", "Vehicle Expenses:Fuel", "Materials - Paper"), which outranks what the vendor list says the vendor usually sells, so a utility's gas bill is gas and a fuel company's courier invoice is courier; then the vendor list; then the most specific word in the row, preferring the vendor's own name to a generic account. A card processor's prefix (PayPal, Square, Toast) no longer decides the row; the merchant after it does. A fuel row that names diesel or petrol is priced as that fuel. The vocabulary now covers fuel stations and fuel cards (US, UK, EU, India), gas utilities including piped gas, metals distributors, staffing agencies, memberships, facility management, food bought for resale and more of the non-emitting accounts. On four real-shaped hand-labelled ledgers (360 rows) this path now prices 95% of spend correctly, up from 77%. And a file covering more than a year is reported for 12 months of it: the latest financial year the file fully covers (from April in India, July in Australia, January elsewhere), or the 12 months you choose; rows outside are set aside and listed.
UK companies' Scope 3 now uses the UK's own published factors: the Defra / University of Leeds spend-based emissions multipliers (kg CO2e per pound, excluding VAT, by SIC product group, in the prices of the year), in place of US EPA factors applied after a currency conversion. Each purchase category is matched to the group of what was bought: a service to its service group, goods to the group that makes them. Spend after 2023, the latest published year, uses 2023's multipliers. The group and year used are printed against each ledger row.
Natural gas everywhere, and electricity outside the US, are now priced the same way as road fuel: the bill is converted to kWh or therms at the official business price for the country and period (EIA monthly for US gas; DESNZ Quarterly Energy Prices for the UK; Eurostat's semi-annual non-household prices for the EU), at a typical small business's size band (electricity 20 to 499 MWh a year, gas below 278 MWh), then multiplied by the combustion factor or the country's grid intensity. Gas had been divided by a US price of $1.00 a therm, about a third of what European businesses pay, and European electricity by prices closer to wholesale than to what a small business is billed. Gas-heavy European ledgers fall 30 to 60% as a result. US electricity keeps its state factors.
Fuel is priced where and when it was bought. A road-fuel or heating-oil purchase is converted to litres (gallons in the US) at the official price for the company's country and the month of the transaction, net of recoverable VAT: EIA retail prices (US), DESNZ weekly road fuel prices (UK), the European Commission Weekly Oil Bulletin (EU) and AIP terminal gate prices (Australia). The quantity is then multiplied by the physical emission factor (DESNZ 2026 for the UK, EPA 2025 elsewhere). Until now every fuel purchase was divided by one US pump price ($3.50 petrol, $4.00 diesel), which overstated European fuel by 1.5 to 2.5 times and drifted with the US market. UK and EU Scope 1 on road fuel falls by about 40%; US Scope 1 on 2025 ledgers rises about 13%, because 2025 pump prices were below the old anchors. Natural gas and propane keep the previous method for now.
Factor Pack v2.6.3. Grid transmission and distribution losses (Scope 3 Category 3) now use each country's published figure where one exists, added as a share of Scope 2 electricity: UK 9.9% (the DESNZ 2026 T&D factor over the generation factor), Ireland 7.8% (Eurostat 2025: 7.2% of production lost), India 21.4% (CEA: 17.6% of electricity available lost). Other countries keep the US eGRID figure, 4.4% (4.2% lost), and the report says so. Losses now apply to purchased electricity only, not steam, heat or cooling. Germany's grid factor moves to the UBA 2025 estimate, 0.352 kg CO2e per kWh (was 0.308). The German state and UK nation-level grid factors are removed: neither had a citable annual source, so the national factor applies.
Factor Pack v2.6.2. Cloud and hosting spend is priced at EPA NAICS 518210 (data processing, hosting and related services), 0.093 kg CO2e per USD. It had been priced at NAICS 334220 (radio, TV and wireless equipment manufacturing), 0.111, which overstated every cloud row by 19 percent.
India factor pack (CEA v21.0, dual-extraction verified): national grid factor updated to the FY 2024-25 weighted average 0.7097 kg CO2 per kWh (CEA publishes CO2 only, stated plainly), spend factor re-derived to 5.93 kg CO2 per USD using PFC's all-India commercial realization of Rs 9.98 per kWh. The 14 previously shipped Indian state factors are REMOVED: CEA publishes one unified-grid factor and no state split, so state-level rows were precision we could not source - Indian reports now use the national factor regardless of state.
Spend-side electricity re-derivation completes the eGRID 2023 rollout: national factor 2.58 to 2.74 kg CO2e per USD (eGRID2023 US average over EIA 2024 final commercial price, 12.75 cents per kWh), all 51 state factors re-derived from eGRID2023 state rates over EIA-861 2024 final state prices (dual-source cross-checked; corrects prior inconsistent price anchors, largest shifts WA +98 percent, TX +70 percent, SD -60 percent), UK spend factor 1.44 to 0.42 (DEFRA 2026 grid over DESNZ 2025 non-domestic price incl CCL) with GB regional values re-anchored to the same denominator. Engine per-report factor stamp moves to Factor Pack v2.6.0.
DEFRA/DESNZ 2026 travel refresh (dual-extraction verified): air per-passenger-km moved to the average-passenger without-RF basis the engine actually matches (no cabin class collected), reflecting DEFRA's sharp 2025 methodology revision - long-haul 0.147 to 0.09043, rail 0.035 to 0.03092, average car 0.168 to 0.16591, UK grid 0.193 to 0.13096 kg CO2e per kWh. EPA Hub 2025 confirmed still current (no 2026 edition exists); one correction: heating oil 10.19 to 10.21 kg CO2 per gallon per 40 CFR Part 98 Table C-1.
eGRID 2023 refresh for the activity path: all 27 US subregion per-kWh factors updated from EPA eGRID2023 rev2 (annual total output CO2e rates, cross-verified against the full data file), Puerto Rico (PRMS) added, US national average 0.374 to 0.3497 kg CO2e per kWh. Spend-based per-USD electricity factors keep the eGRID 2022 derivation until re-anchored with matched EIA prices.
Spend-based factors recalibrated to EPA Supply Chain GHG Emission Factors v1.3 (2022 USD), correcting prior stored values. Added Scope 3 Category 8/9/12 factor coverage with primary-source citations, per-factor derivation checks, and an automated regression test suite wired into CI.
NAICS sector expansion · 27 → 66 emission types. International grid-factor parity (EU, India, Japan, UK). Notes-column inference, data-salvage summary, per-row inference decision log. TTF font name-table fix for PDF embedding; font-scrub pass removes standard-14 declarations from PDF resource dictionaries.
Phase 5 intake layer: encoding detection (Windows-1252 + UTF-8), EU decimal convention, parens-negative format, localized header map (German Lieferant → vendor, Betrag → amount).
Phase 4 classifier upgrades: rapidfuzz fuzzy matching, AP transaction-prefix regex (AMZ*, SQ*, PAYPAL*, SHELL-OIL, MSFT AZURE), expanded KNOWN_VENDORS with SaaS/utilities/lodging. 84% high-confidence on messy AP data without Claude.
Factor table refresh: USEEIO v2.2 (Oct 2024 publication), eGRID 2022 subregional mapping, IPCC refrigerant GWP100 values (AR4).
Initial public engine. GHG Protocol Corporate Standard + Scope 3 Standard aligned. EPA USEEIO industry factors.
Read the specimen.
The disclosure-aligned report is the methodology in practice: every factor cited, every credit netted, every row traceable to its source.
Generate a draft · from $20