Reconciled to the exact euro: Provenship against the EMSA 2023 public dataset

A maritime compliance engine is only as trustworthy as the proof it can offer the verifier sitting across the table from the owner. Provenship's proof is a euro figure produced two independent ways from the same regulator-published source data, reconciled to ±0.0000%. This note documents the reconciliation: how the source data was ingested, how the calculation was structured, and — equally important — what was kept out of scope for v1.

Source: EMSA THETIS-MRV 2023

The European Maritime Safety Agency publishes annual emission reports under EU MRV Regulation 2015/757. The 2023 dataset (downloaded from EMSA's THETIS-MRV portal) covers 12,829 vessels above 5,000 GT calling at EEA ports. After filtering for entries that carry fuel-consumption data (some vessels in the dataset are reported but ungauged), 12,637 vessel-years remain.

Every input to the engine comes from this single CSV. We deliberately did not enrich, infer, or fill missing rows. Where the regulator's published source is silent, the engine refuses rather than approximates — see the dedicated note on refusal as a first-class outcome.

Two paths, one answer

The engine ran the dataset two ways:

• Calculation-only path. ProofCli loads the CSV in-process, applies the FuelEU Annex IV Part B formula chain (energy → gCO2eq → tonnes-VLSFO-equivalent → EUR) and aggregates per-fleet.
• Full-stack path. The same rows are written through the production handlers under TenantGuard transactions, persisted to Postgres, and read back through the same query layer the production API exposes. The penalty figure is then aggregated through the same SQL the operator dashboard uses.

Both paths produced €535,884,359 as the aggregate projected 2026 FuelEU Maritime penalty. The delta between the two paths was 0.0000%. Total runtime: 170 seconds (53s ingest, 116s projection).

Nine regulator-derived JSON fixtures

Beyond the EMSA reconciliation, every formula in the engine is pinned to a JSON fixture derived from a regulator citation. The fixtures cover the FuelEU Annex II carbon-intensity defaults, the Annex IV penalty formula, the EU ETS phase-in schedule under Directive 2003/87/EC, and the CII reference lines from MEPC.354(78). The 67/67 automated tests passing is the smaller, finer-grained sibling of the dataset-level reconciliation.

What was out of scope for v1

Honest disclosure: the v1 reconciliation does not cover everything an operator needs.

• CII verdicts. EMSA's published dataset does not include the IMO Carbon Intensity Indicator rating per vessel; CII reconciliation requires a separate IHS Markit corridor lookup we have not yet licensed. CII calculations are implemented but unverified against a third-party source.
• Multi-fuel Bunker Delivery Note (BDN) splits. The EMSA dataset records consumption by primary fuel; vessels that switched between HFO and biofuel within a reporting period are summarised. The engine accepts multi-fuel BDNs at run time but the EMSA reconciliation did not test that path.
• Pool agreements. FuelEU's pooling provisions reshape the penalty figure for operators that join a compliance pool. v1 computes the standalone penalty only.

What this buys the operator

An operator who hands Provenship their bunker delivery notes and voyage records gets back the same euro figure the verifier would compute from the same inputs, with the audit trail to defend it. The reconciliation report — formula derivations, fixture sources, EMSA appendix, engine version — is available on request.