Greenhouse gas inventory workpaper (GHG Protocol Scope 1 / 2 / 3)

Structure (9 sections)

1. 1. Organizational boundary: consolidation approach and entity list

   First fix and record the consolidation approach: equity share, financial control or operational control. Once chosen it must be applied consistently across Scope 1, Scope 2 and Scope 3 — the Scope 3 Standard requires that consistency explicitly. Then list every entity: name, shareholding, control relationship, whether included, inclusion percentage and the basis. Three boundary questions recur for Chinese AI companies: (1) whether VIE-structure entities are consolidated; (2) whether a data hall built with a cloud provider or a colocated rack counts as own operations or purchased services; and (3) whether a jointly-owned compute centre is treated on equity share or control. Wherever judgement is applied, write down the reason and keep the approval record.

   Evidence to attach

   • · Note on the consolidation approach chosen, with management approval
   • · Entity list (name, shareholding, control relationship, inclusion, inclusion percentage, basis)
   • · Reconciliation against the financial statement consolidation scope
   • · Extracts of colocation or joint-build contracts used to assess where control sits

   Common pitfalls

   • · Using operational control for Scope 1 but equity share for Scope 3, producing an inconsistent basis that cannot be recalculated
   • · Excluding colocated facilities wholesale when operational control in fact rests with you
   • · Listing only PRC entities and omitting overseas sales subsidiaries and R&D centres

2. 2. Scope 1: identifying direct emission sources and activity data

   Build a ledger source by source: stationary combustion (on-site diesel gensets, standby generator test runs, gas boilers); mobile combustion (company vehicles and test fleets — the road-test fleet of an autonomous driving or embodied-AI company is a commonly overlooked major item); process emissions (etching and cleaning gases for semiconductor and hardware manufacturers); and fugitive emissions (refrigerant leakage from data hall and office cooling; where you own the data hall or its electrical rooms, also SF6 leakage from medium- and high-voltage gas-insulated switchgear — SF6 sits in switchgear, not in UPS battery rooms, whose emission risk is hydrogen and fire rather than greenhouse gases). For each, record activity data, unit, data source (fuel invoices, maintenance records, procurement ledgers) and the period covered. Refrigerant leakage is the most commonly missed: approximate the top-up quantity as opening stock + purchases − closing stock, but back out the initial charge of newly installed equipment and the refrigerant recovered from decommissioned units (neither is a leak), and keep the maintenance records.

   Evidence to attach

   • · Inventory of stationary and mobile combustion sources with fuel consumption ledgers (invoices, fuel cards, fleet system exports)
   • · Refrigerant ledger (type, source of the GWP value, top-up quantities, maintenance records)
   • · Generator test-run logs and fuel issue slips
   • · Process gas purchase and usage records where applicable

   Common pitfalls

   • · Counting only office vehicles while omitting the road-test fleet and outsourced transport vehicles under your control
   • · Treating refrigerant leakage as zero across the board, which cannot be defended in assurance
   • · Not labelling which IPCC assessment report the GWP values come from, making figures non-comparable year to year

3. 3. Scope 2: location-based and market-based dual reporting

   The Scope 2 Guidance requires companies with operations in markets that provide contractual instruments to calculate and separately label both a location-based and a market-based result (dual reporting). The location-based method uses the average factor of the grid where consumption occurs (in China, the power CO2 emission factors published by the Ministry of Ecology and Environment; in the US, EPA eGRID and the Emission Factors Hub; in the UK, the DESNZ conversion factors). The market-based method uses contractual instruments (certificates, direct power purchase agreements, supplier-specific rates) that must meet the Scope 2 Quality Criteria, falling back to a residual mix factor where no qualifying contractual information exists. The workpaper should list, site by site: consumption, grid region, location-based factor and its data year, the type and volume of contractual instruments held, and the market-based factor and its basis. Compute-intensive companies should also separate data centres from offices, because customer questionnaires often ask only about the data-centre figure.

   Evidence to attach

   • · Site-level electricity ledgers (utility bills, sub-meter readings, colocation provider monthly statements)
   • · Record of location-based factor sources and data years (MEE announcements, eGRID, DESNZ)
   • · Evidence for contractual instruments — certificates, direct PPAs, supplier emission rate statements — with retirement records
   • · Decision record for market-based factor selection, including the Scope 2 Quality Criteria assessment
   • · Dual-reporting comparison table (location-based vs market-based, by site and in total)

   Common pitfalls

   • · Reporting a single Scope 2 number without stating the method — both customers and exchanges will ask
   • · Disclosing only the market-based result after applying certificates and hiding the location-based result
   • · Having only cost invoices from a colocation provider, estimating consumption from rack count without disclosing the estimation method
   • · Applying the same certificates in both the parent and subsidiary disclosures, causing double counting

4. 4. Scope 3: screening the 15 categories and setting the boundary

   Screen all 15 categories in Table 5.3 of the Scope 3 Standard for relevance, recording for each: whether you have that activity, a first-pass magnitude estimate, data availability, and the include/exclude conclusion with its reason. The standard requires reporting by category and permits exclusions provided they are disclosed and justified. For Chinese AI companies the usual leaders are Category 1 (cloud and compute procurement, outsourced data labelling, SaaS subscriptions), Category 2 (GPU servers, test equipment, robot units), Categories 6 and 7 (cross-border travel and commuting), Category 11 (inference energy in the use phase of sold products, relevant to both model API and device companies) and Category 12 (hardware end-of-life). The screening conclusion should be signed off by management rather than decided unilaterally by one analyst.

   Evidence to attach

   • · A 15-category screening sheet (per category: activity present, magnitude estimate, data availability, include/exclude conclusion and reason)
   • · Management approval of the screening conclusion
   • · Finance export of procurement spend mapped to categories (for spend-based first-pass screening)
   • · Written justification for excluded categories and the plan to bring them in

   Common pitfalls

   • · Disclosing only the easy categories (travel and commuting) while excluding the largest — compute procurement and the product use phase
   • · Marking a category "not applicable" when the activity plainly exists
   • · Failing to break down by category and reporting only a Scope 3 total
   • · Ignoring the Table 5.4 minimum boundaries, e.g. counting only transport in Category 1 rather than all cradle-to-gate emissions

5. 5. Scope 3: choosing calculation methods and grading data quality

   For each included category, state the method used (supplier-specific, average-data, spend-based or hybrid) and why, and grade the data quality. The principle: push the largest categories towards supplier-specific data, and leave small categories on a spend-based first estimate — but label the method tier. Two practical points for AI companies. Cloud and compute procurement (Category 1) should first seek account-level carbon reports from the cloud provider; failing that, estimate as consumption × grid factor × PUE and state where the PUE value comes from. For Category 11 inference energy, spell out the unit assumptions (energy per request, annual request volume, device service life) — these assumptions attract the most audit challenge and drift most easily between years.

   Evidence to attach

   • · Per-category method selection table (method, rationale, data quality grade, owner)
   • · Carbon data reports or written responses from cloud providers and contract manufacturers (including records of refusals)
   • · Basis for key assumptions such as PUE, energy per inference and device service life
   • · Version of the spend-based intensity factor database and the mapping table used
   • · Data quality improvement plan listing categories to be upgraded next year

   Common pitfalls

   • · Using the spend-based method everywhere while describing the result as "based on supplier data"
   • · Quietly tuning key assumptions (PUE, energy per inference) year over year to flatter the numbers
   • · Recording a data quality grade with no stated grading criteria, so it cannot be reviewed
   • · Citing a cloud provider carbon report without checking that the accounts and regions it covers match your own

6. 6. Emission factor register and version control

   Maintain one register covering every factor used: name, value, unit, applicable region, data year (not publication year), publisher, source link, basis (location-based / market-based / residual mix), whether non-CO2 gases are included, and where it is used in the inventory. Factor updates are the first killer of year-on-year comparability: when an official body publishes a new data year, decide whether to recalculate prior years and record the decision and its reason in the register. China grid factors are published by the Ministry of Ecology and Environment by data year and at both national and regional level, and must be matched to where the electricity is consumed; multinational operations should register the local official factor for each overseas site separately.

   Evidence to attach

   • · Emission factor register (data year, basis, source link, where used)
   • · Factor version change log and the decision on whether prior years were recalculated
   • · Archived copies of the official factor publications (PDF or page snapshot)
   • · Note on the local official factors used for overseas sites

   Common pitfalls

   • · Labelling the publication year as the data year, which shifts your figures a year out of step with peers
   • · Applying the China national grid factor to every site worldwide
   • · Updating the factor for the current year only, without recalculating or explaining prior years, so the trend chart is distorted

7. 7. Base year and recalculation policy

   Fix in writing the base year, base-year emissions, the triggers for recalculation and the recalculation threshold (typically: structural change such as acquisitions or divestments, changes in methodology or factors, and discovery of prior-period errors). Express the threshold quantitatively (for example, recalculate once the cumulative effect exceeds a stated percentage of base-year emissions) and then apply it consistently. The moves Chinese AI companies make frequently — setting up an overseas subsidiary, acquiring compute assets, divesting a hardware line — are all structural changes requiring base-year recalculation; without it, the decline in the emissions curve will be read by auditors and investors as a boundary change rather than a real reduction.

   Evidence to attach

   • · Base-year policy text (base year, base value, triggers, quantitative threshold)
   • · List of structural changes in the period with the recalculation workings
   • · Before/after comparison and the effect on target progress
   • · Management approval of the recalculation decision

   Common pitfalls

   • · No written recalculation policy, so the decision is taken by feel each year
   • · Adding an acquired entity in the year of acquisition without recalculating the base year, distorting the year-on-year comparison
   • · Recalculating without disclosing the quantified effect, leaving readers unable to understand why the numbers moved

8. 8. Inventory quality management and calculation trail

   Put in place a reviewable trail: for every calculation unit keep the four-part record of activity data → factor → formula → result, naming the data provider, the reviewer and the review date; set cross-checks on totals (electricity consumption against electricity spend, travel mileage against travel expense claims); and require a written explanation for any anomalous movement beyond a set threshold. This section drives the cost of a future limited assurance engagement directly — with no trail, the practitioner can only expand sampling or qualify the conclusion.

   Evidence to attach

   • · Calculation workpapers (activity data, factor, formula, result, preparer, reviewer, review date per unit)
   • · Cross-check table on totals (energy vs spend, mileage vs reimbursements)
   • · Anomaly explanation sheet
   • · Internal review sign-offs and the issue remediation log

   Common pitfalls

   • · Keeping every calculation in one unversioned spreadsheet, so an overwritten formula cannot be reconstructed
   • · Retaining results without intermediate steps, so the work is lost when staff change
   • · Leaving anomalies unexplained and then facing line-by-line analyst questions after publication

9. 9. Hand-off package for assurance and external disclosure

   Package the workpapers so they can be handed straight to an assurance practitioner or a customer: the inventory summary (by scope, category, entity and site), the methodology note, the emission factor register, the significant assumptions and uncertainty note, the boundary change and recalculation note, and an index of supporting documents. Also prepare a customer-facing summary: most enterprise customers want only three figures (Scope 1, Scope 2 location-based and market-based, and the portion of Scope 3 relevant to them) plus a sentence on method — turning that into a standard response materially reduces the marginal cost of every questionnaire.

   Evidence to attach

   • · Inventory summary by scope, category, entity and site
   • · Methodology note (boundary, methods, factors, assumptions, limitations)
   • · Index of supporting documents and where they are stored
   • · Customer-facing summary (a version cleared for external distribution)
   • · Tie-out sheet reconciling the workpapers to the published disclosure text

   Common pitfalls

   • · Published figures that do not tie back to the workpaper summary, usually because the disclosure text was hand-edited late
   • · Giving customers a carbon figure on a different basis from the one filed with the exchange, without explaining the difference
   • · Supporting documents scattered across personal machines and unavailable within the agreed assurance deadline