How Hemera works.

For every activity record we assign the highest tier the evidence supports. The evaluation runs top to bottom; the first tier whose evidence requirements are satisfied wins. Tier downgrades are typed events with a documented reason. The platform will not let an analyst claim Tier 1 when the evidence supports Tier 4.

Each tier carries an explicit data-quality score derived from a five-axis pedigree matrix: reliability, completeness, temporal correlation, geographical correlation, and technological correlation. Those scores feed directly into a per-row geometric standard deviation (GSD), which propagates upward into the workspace-level confidence interval.

Choosing the right tier for a supplier line item is the single largest source of pedigree improvement in a typical workspace. Hemera runs a structured eight-layer research method on every supplier the first time we encounter them, and reuses the resulting profile for subsequent encounters until its annual review date.

The layers are evaluated in order. The first layer that yields data above the workspace’s configured tier threshold ends the search; subsequent layers are skipped unless the analyst explicitly continues for a higher-confidence answer.

1. Supplier name and spend.Normalise the row’s supplier string to a working identifier; preserve the original for traceability.
2. Entity resolution. Look up the supplier in Companies House (UK) or the equivalent registry; attach a stable entity id. Resolution failures flag for admin review, never silent defaults.
3. Sector and sub-sector classification. Assign a NACE sector and Hemera sub-sector. Drives the Tier 3 Exiobase factor and informs the DEFRA match.
4. Primary-data probe. Search for a verified PCF, a CDP response, or comparable third-party-assured emissions disclosure. A verified PCF unlocks Tier 1.
5. Parent and group identification.Establish the supplier’s corporate group; a subsidiary may inherit the parent’s PCF only when the parent attests to subsidiary inclusion.
6. Country of origin.Determine where the goods or services are actually produced. For trading entities, that’s the production country, not the invoicing country.
7. DEFRA category match. With sector, sub-sector, and country in hand, attempt a DEFRA factor match. Success moves the line from Tier 3 to Tier 2.
8. Exiobase fallback. Apply the UK sector intensity from Exiobase at Tier 3. Always available; the default when none of layers 4 to 7 yield a higher answer.

The output of the eight layers is a usable supplier profile: sector, country, evidence trail, and a recommended tier, which the platform feeds back into the four-tier data hierarchy above.

Every figure in a Hemera workspace is anchored to four things, and only the platform can write any of them:

• A versioned emission factor: the factor library version that was current at calculation time is pinned to the row, so historical periods reproduce exactly.
• A hash-chained audit event: every write is signed against the prior event hash. Mutating any row deterministically breaks the chain. The platform refuses UPDATE and DELETE against audit-tracked tables outside the sanctioned service layer.
• The analyst who approved it, tagged with a typed provenance_actor: analyst, ai_manual, ai_automated, or rule_based.
• Speciated greenhouse gases at row level: CO₂, CH₄, N₂O, HFCs, PFCs, SF₆, NF₃ separately, not just CO₂e. Frameworks that ask for speciation get the real numbers.

A re-run of last year’s report next year produces the identical figure, to the gram. Verifiers can re-execute the audit pack and confirm that replay independently, on their own machine.

One canonical inventory; many disclosure shapes. Each framework is a projection of the same underlying ledger, not a parallel calculation. Here’s what we ship today:

The five-axis pedigree score on each record maps to a per-row GSD via a published lookup. Independent rows propagate variances in quadrature; non-linear aggregations (intensity ratios, percentages of varying denominators) are run through a Monte Carlo with a deterministic, methodology-pinned seed for reproducibility.

Every workspace headline figure carries a 95% confidence interval (lower bound, point estimate, upper bound), surfaced on the cover of the methodology report and on every framework projection. Most platforms hide uncertainty. We don’t. Decisions calibrated against the wrong number are worse than slow decisions.

We are most confident about the comparisons: what your biggest source is, which data upgrade helps most, whether a change is big enough to show. The exact width of the range is our best estimate, built conservatively, and it sharpens as real data replaces assumptions.

The range also polices claims. When a claimed reduction is smaller than the range around your number, the software says so before someone else does.

A published methodology report is signed by the workspace owner. The signature is their own declaration that the figures are theirs. Hemera staff then release the report. Hemera does not review, verify, or assure the figures, and nothing in the product should be read as saying otherwise.

The sign-off snapshot captures the signer and the moment of signing. The historical signature remains valid even if the signer’s details later change, because the snapshot is immutable. The signature log itself is append-only.

What Hemera brings is the method: the boundary rules, the factor hierarchy, the uncertainty model, and the audit trail. What the report says about your emissions rests on the data you give us.

For every published report Hemera generates a locked-format ZIP. It is everything an external verifier needs: on disk, offline, with no Hemera dependency.

CBAM-regulated workspaces also receive per-product embedded-emissions files in the same pack.

The bundled verify_pack.py is stdlib-only Python. It runs on any machine without pip install. It performs three independent checks:

1. SHA-256 of every file in the pack against the manifest.
2. Replay of hash_of_inputsfrom the input snapshot, matching the manifest’s expected hash.
3. Audit-chain prev_hashlinkage walk: the chain holds end-to-end or it doesn’t.

Exit code 0 prints OK with the format version, artifact id, and hash prefix.

An honest list. If a buyer asks “does Hemera do X?” and X is below the line, the answer is “not yet” or “not in the next 90 days”.

• Full SAP, Oracle, NetSuite ERP integrations. They land as customer demand pulls them in. CSV upload + AI utility-bill extraction cover the interim.
• Direct utility API connectors. Octopus, British Gas, etc. The AI utility-bill extractor (analyst-confirmed before any write) is the path today.
• Monte Carlo over the supplier graph. Per-row GSD propagates to workspace level today; correlated supplier-graph rollup is research, not a production feature.
• Abatement cost modelling. What-if reduction scenarios (cut a category, switch a supplier or factor source) are built in and computed against your current inventory, alongside SBTi-aligned targets. We do not yet model time-phased intervention pathways or marginal abatement cost curves.
• Auditor multi-workspace portfolio view. One workspace at a time today.

Contains public sector information licensed under the Open Government Licence v3.0 (https://www.nationalarchives.gov.uk/doc/open-government-licence/version/3/). Emission and EEIO factors are Crown copyright, DESNZ / DEFRA; exchange rates are Crown copyright, HMRC. Company records used for entity resolution are from Companies House, also under the Open Government Licence v3.0.

Tier 3 spend-based intensities are derived from EXIOBASE v3.8.2, Stadler, K. et al. (2021), DOI 10.5281/zenodo.5589597, used under the Creative Commons Attribution-ShareAlike 4.0 International licence (https://creativecommons.org/licenses/by-sa/4.0/). Hemera converts the published monetary intensities to a per-GBP basis and maps them to its own sector taxonomy. Those adaptations are made available under the same ShareAlike terms on request.