AI & Methodology Disclosure

HemeraScope calculates carbon footprints using DEFRA GHG Conversion Factors (published annually by the UK Department for Environment, Food & Rural Affairs). We apply a cascading precision model:

• Level 1 — Supplier-specific: Where a supplier provides verified emissions data (e.g. via CDP, SBTi, or direct disclosure), we use their reported figures. This is the highest-fidelity tier.
• Level 2 — Activity-based: Where specific activity data is available (e.g. kWh of electricity, litres of fuel, tonnes of waste), we apply the corresponding DEFRA conversion factor.
• Level 4 — EEIO spend-based:Where only spend data is available, we apply environmentally extended input–output (EEIO) emission factors by sector and spend amount. This is the lowest-fidelity tier and is clearly flagged in reports.

Every carbon figure in HemeraScope includes an uncertainty range expressed as a geometric standard deviation (GSD) with a 95% confidence interval. Data quality is scored using a Pedigree Matrix aligned with ISO 14044 requirements, assessing reliability, completeness, temporal correlation, geographical correlation, and technological correlation.

HemeraScope builds supplier risk profiles using a 13-layer enrichment protocol that draws on publicly available registries and datasets, including:

• Companies House (company status, filing history, directors, SIC codes)
• Environment Agency (environmental permits, pollution incidents, enforcement actions)
• Health and Safety Executive (HSE notices, prosecutions)
• ICO Register of Data Controllers
• Modern Slavery Statement Registry
• Contracts Finder (public procurement history)
• Science Based Targets initiative (SBTi) commitments
• CDP disclosure status

Findings from public registries are deterministic— they are direct lookups against authoritative sources and are presented as factual records with source attribution. AI-assisted risk analysis is applied on top of these deterministic findings and is always clearly distinguished in reports.

HemeraScope uses Claude (developed by Anthropic, PBC) for the following purposes:

• Transaction classification: Categorising spend line items into emissions categories for carbon calculation
• Supplier risk analysis: Synthesising findings from multiple public registries into coherent risk narratives
• Report generation: Producing client-language summaries and explanatory text for deliverables

Important safeguards:

• AI outputs are always reviewed by a human analyst before delivery to the client. We do not send AI-generated content directly to clients without review.
• We do not fine-tune AI models on client data. Transaction descriptions and supplier names (with no personal identifiable information) are sent to the Anthropic API for processing; this data is not used by Anthropic to train their models, per our data processing agreement with Anthropic.
• AI-generated analysis is clearly labelled in reports so clients can distinguish between deterministic findings and AI-assisted interpretation.

Our quality assurance process is inspired by ISO 19011 audit sampling principles:

• Stratified sampling: A representative sample of transactions and supplier records is manually reviewed for every engagement, with sampling rates weighted by risk and value
• Hard gate threshold: If the error rate in any sample exceeds 5%, the entire batch is returned for reclassification before proceeding
• Source traceability:Every finding in a HemeraScope report is traceable to its source — whether a public registry record, a DEFRA conversion factor, a client-provided data point, or an AI-generated analysis

HemeraScope is aligned with the following standards and frameworks:

• GHG Protocol Corporate Standard— for scope 1, 2, and 3 emissions categorisation and reporting
• ISO 14064-1— for greenhouse gas quantification and reporting at the organisation level
• TCFD recommendations— for climate-related financial disclosure alignment

Note: We are aligned with these standards but not independently certified against them. HemeraScope outputs are estimates prepared using recognised methodologies and are not a substitute for independent verification or third-party assurance where required by a specific regulatory regime.

We update this page when our methodology, AI usage, or quality control processes change materially. The “Last updated” date at the top reflects the most recent revision.