Methodology

1. Frameworks

The calculator follows the GHG Protocol's Scope 2 (location-based) method for operational electricity, plus Scope 3 Category 11 for amortized embodied (manufacturing) emissions. Service-level energy coefficients come from Cloud Carbon Footprint, the de-facto open-source standard. Grid carbon intensities default to IEA 2023 medians and are overridden with real-time Electricity Maps data when available.

2. Core formula & grid intensity

For each service category, we compute operational energy in kWh, multiply by the regional grid intensity at calculation time, then add embodied (manufacturing) carbon. The total emissions for a vendor is the sum across all services across all connected cloud accounts.

Energy_kWh   = usage × kWh_per_unit(service) × PUE(provider)
Scope_2_kg   = Energy_kWh × grid_intensity_g/kWh / 1000
Scope_3_kg   = embodied_factor_per_unit × usage_units      (v3, see §4)
Total_kg     = Scope_2_kg + Scope_3_embodied_kg + Scope_3_transport_kg

Grid intensity resolution (v3 Pass 2)

Annual averages misrepresent emissions because intensity varies 5–20× across the day. us-east-1 swings between ~250 g/kWh at solar peak and ~520 g/kWh on overnight gas peakers; eu-west-3 swings from 18 g/kWh (nuclear baseload) to 200+ g/kWh on winter peak imports. Verdex therefore queries Electricity Maps for the trailing-24h hourly series at calculation time and applies the time-weighted mean (trapezoidal integration over the hourly samples). The series is cached in Firestore for 1 hour per zone so concurrent verify runs share one upstream call.

region                  → Electricity Maps zone (lib/grid-intensity.ts)
zone                    → /v3/carbon-intensity/history (24 hourly samples)
hours[]                 → time-weighted mean (gCO2e/kWh)
fingerprint = sha256(canonical_json({zone, hours}))     ← in proof.json

fallback chain:
  no API key / unknown zone / upstream error
    → REGION_INTENSITY[region]   (IEA 2023 annual mean)
    → FALLBACK_REGION_INTENSITY  (world average, 400 g/kWh)

Every score document records the actual intensity that was applied (region_intensity_g_per_kwh), its provenance (region_intensity_source: hourly_trailing_24h / cache_hit / static_annual_average), and a SHA-256 fingerprint over the hourly series so a buyer's auditor can replay the exact mean.

Honest caveat. v3 ships hourly intensity, not hourly usage matching. Usage is assumed flat across the day — mathematically equivalent to multiplying monthly kWh by the time-weighted mean. True hourly matching needs hourly usage data (AWS Cost Explorer is monthly-only; Azure Carbon Optimization and GCP BigQuery export are month-grained). That lands in Pass 4 when the AWS Sustainability service GA (June 2026) adds per-hour exports. The GHG Protocol's 2026 Scope 2 consultation is moving toward hourly matching as a requirement for market-based claims by 2027.

3. Energy coefficients

4. Embodied carbon — absolute per-unit factors (Scope 3 Cat 11)

v3 replaces the v2 "multiplier on operational" model with absolute per-unit factors. Why: as grids decarbonize, embodied carbon's share of lifetime data-center emissions has grown from ~25% to 40–50%. In clean-grid regions (eu-north-1 at 8 g/kWh) a multiplier-of-operational under-represents embodied by an order of magnitude. Absolute per-unit factors stay correct regardless of grid. Numbers derive from iMasons Climate Accord server LCA disclosures (~922 kg CO₂e per dual-socket server) amortized over a 4-year service life and split by vCPU + memory density.

Compute — kg CO₂e per instance-hour

Storage, transfer, serverless, builds

Fallback: when instance type can't be resolved (e.g., AWS Cost Explorer spend without per-SKU breakdown), the v2 multiplier model is used and the score is tagged as spend_to_kg_fallback with a ±35% confidence band.

5. AWS Cost Explorer fallback factors

When granular usage isn't available (no CUR, no per-instance metrics), we fall back to converting AWS spend into emissions using category-specific factors. These are derived from Cloud Carbon Footprint coefficients combined with AWS list pricing for the dominant SKU in each category. CUR ingestion gives ~5–10× tighter accuracy and is on the registry roadmap.

6. Score & tier

ratio  = total_kg / size_band_benchmark
score  = clamp(round(1000 × (1 − ratio)), 0, 1000)

Gold     800–1000   (≤ 20% of benchmark)
Silver   600–799    (20–40% of benchmark)
Bronze   400–599    (40–60% of benchmark)
Tracked    0–399    (> 60% of benchmark)

Score is a ratio against a CCF reference deployment at the vendor's size band — not a percentile rank. Each band's benchmark is a deterministic deployment (vCPU + storage + egress mix) run through the same methodology described here. The registry profile says "your emissions are X% of the size-band reference," not "you beat X% of vendors." Once the registry has ≥50 published vendors per band, this benchmark becomes a fallback and a true percentile fromregistry_snapshots takes over.

6a. Market-based Scope 2 (dual-reporting)

The GHG Protocol's Scope 2 standard requires dual reporting: a location-based number (what was physically emitted on the grid the data centre draws from) and a market-based number (after applying the contractual REC/PPA/CFE claims the cloud provider has made). CSRD and the SEC climate-disclosure rules both expect both numbers. Every Verdex score document therefore carries:

scope_2_kg                 ← location-based (physical grid mix)
scope_2_market_kg          ← location_kg × market_factor.factor
scope_2_market_basis: {
  basis: "provider_100pct_match" | "provider_partial_cfe"
       | "inherited" | "no_claim" | "self_declared_ppa",
  factor: 0.0–1.0,
  source: <citation URL>,
  year:   <year of disclosure>,
  note:   <caveat>
}

The headline total_kg always uses the location-based number — that is the auditable physical footprint. The market number is reported alongside so a buyer can see how much of a "clean" claim depends on REC accounting vs. actually-clean grids.

Provider claims (2024 / 2023)

Honest caveat. An annual 100% REC match does not mean the data centre is running on 100% renewable power at any given hour — it means the provider purchased a matching quantity of renewable energy certificates over the year. Only GCP currently publishes hourly carbon-free-energy data per region. The GHG Protocol's 2026 Scope 2 consultation is moving toward hourly matching as a 2027 requirement for market-based claims; vendors will then need to disclose at the hourly grain.

7. Confidence intervals

v3 reports a +/-% confidence interval alongside the headline number. Every score document carries aconfidence_interval_pct field picked from the table below based on how the underlying data was obtained. CSRD reporting requires uncertainty disclosure; the interval is shown as 45.4 kg ± 15%on the registry profile.

The legacy accuracy_level enum (high / medium / low / self-reported) is still emitted on every score document for back-compat with old consumers.

8. Per-scope provenance

Each score document includes adata_sources_by_scope field recording exactly which method produced each GHG scope. Auditors should not need to guess which input fed which output:

data_sources_by_scope: {
  scope_2_location: [
    { provider: "aws", source: "cost_explorer",
      method: "spend_to_kg_per_category",
      hash_of_response: "sha256:..." }
  ],
  scope_3_embodied: [
    { provider: "imasons_meta_lca", source: "embodied_kg_per_instance_hour",
      method: "absolute_per_unit" }
  ],
  scope_3_transport: [ ... ]
}

9. What this methodology does not include

For honesty with vendors and auditors, the following are explicitlyout of scope for the current methodology version:

• Scope 1 — direct emissions from owned vehicles or fuel combustion.
• Scope 3 categories other than Cat 11 (purchased goods, business travel, etc.).
• Third-party auditor signature on the calculation. Verdex signs the proof JSON itself; an external attestation is on the roadmap and not present today.
• The full calculation breakdown lives in the signed proof JSON. Verification is off-chain — via the certificate's Ed25519 signature and its Certificate Key.
• Renewable Energy Certificates (RECs) or market-based Scope 2 — Verdex uses location-based grid intensity only. Market-based Scope 2 is planned (Pass 3 of the methodology roadmap) alongside hourly grid matching.
• Hourly grid intensity matching — v3 still uses long-run regional averages; weighted hourly intensity × hourly usage is planned (Pass 2) ahead of the GHG Protocol's pending 2027 standard.

10. Reproducing a vendor's score

Each vendor's public profile exposes a signedproof.json containing the full per-service breakdown, themethodology_version, and a SHA-256 fingerprint of each upstream API response we used. To reproduce: fetch the proof, look up the constants on this page for the named version, multiply the usage by the cited factors, and verify your sum matches the proof's total.