HarborProof: the EU-MRV fleet under FuelEU Maritime's penalty formula

FuelEU Maritime (Regulation (EU) 2023/1805) caps the well-to-wake GHG intensity of the energy a ship uses: the reference value is 91,16 gCO2eq/MJ and the limit steps down -2% from 2025, -6% from 2030, -14,5% from 2035, -31%, -62%, -80% (Art. 4(2)). The compliance balance is (GHGIE_target − GHGIE_actual) × energy (Annex IV A) and the penalty is |CB| / (GHGIE_actual × 41 000) × EUR 2 400 (Annex IV B) — a closed-form function of numbers a company must already report. Its first reporting period is 2025, and the penalty for it fell due by 30 June 2026 (Art. 23, read from the pinned text): it is in force, not coming.

Every constant used here is transcribed from the pinned Official Journal bytes — corpus/sources/fueleu_reg-2023-1805_oj-l234_2023-09-22.pdf, sha256 492df4f7… — and the GWP factors from the pinned RED II Annex V C(4) (CO2 1 · CH4 25 · N2O 298). The instrument's battery re-derives the limits, a hand-computed penalty and a hand-computed flip threshold from those constants at every build, and a forged intensity one part in 10⁹ over the limit must flip the verdict — no tolerance window exists for an optimizer to land in.

The activity record is the EU-MRV public emission report for period 2025 — 17,045 ships' verifier-checked annual totals: fuel consumed, CO2eq emitted, and the voyage split (intra-EU / inbound / outbound / in-port) that determines FuelEU scope. Version 45, generated 28 August 2026, fetched from the THETIS-MRV public API and pinned: 1b139b29222dd58c…. The build re-hashes the bytes and refuses on drift; the extraction is stdlib Python, keeps every figure as the exact decimal string the registry printed, and is rerunnable in one command.

No ship's operation is modeled here: no speed curves, no engine assumptions, no AIS. The energy is the ship's own reported fuel; the scope share is the ship's own reported voyage split. What the registry does not say — which fuel, exactly — enters as a box under two stated hypotheses (§3), and the box is the honest price of the record's opacity.

The fossil-oil hypothesis. A decided ship is assumed to burn some mix of the regulation's three pinned oil rows (HFO, LFO, MGO): LCV in [0,0405, 0,0427] MJ/g, well-to-tank in [13,2, 14,4] gCO2eq/MJ. Its tank-to-wake intensity is not assumed at all — it is the ship's own reported CO2eq divided by its own reported fuel. A plausibility screen (prune-only, like every screen in this machine) admits the hypothesis only when some mix of the pinned rows reproduces that ratio — [3,1688, 3,2609] g/g, where the pinned oil rows themselves live; LNG burners, methanol, heavy bio blends and inconsistent reports fall outside and are counted, never decided. 2,385 ships sit there today.

The box is constrained by the ship's own numbers, then decided whole. The reported CO2eq/fuel ratio pins the feasible oil mixes to a one-parameter family, and the intensity extremes of that family sit at two-fuel vertex blends — computed exactly, so each box is tight: it contains every mix consistent with the observation and nothing else. COMPLIANT means every point of the box clears the limit; NON-COMPLIANT means every point exceeds it; a straddling box is NEEDS DATA, never guessed (0 today — the tightened boxes decide everything the screen admits). The undecided mass lives in the 2,385 ships OUTSIDE the oil screen — LNG, methanol, bio blends: precisely the candidates for actual compliance, and precisely the rows the public record cannot yet decide. The per-fuel breakdown FuelEU already makes companies compute would decide every one; NEEDS DATA is this page measuring that opacity. Penalty sums use directed rounding (floors on the floor, ceilings on the ceiling), so the stated exposure range is a theorem under the stated hypotheses, not an estimate.

What this is not. FuelEU compliance is assessed per company with banking, borrowing and pooling; these figures are the raw per-ship Annex IV arithmetic before those flexibilities, under the stated boxes — a mathematically certified counterfactual, not a compliance determination, not a verification statement in the class-society sense, and not legal advice. Scope shares inherit the stated uniform-fuel-mix-across-voyages hypothesis; ice-class and OPS derogations are not applied. Every one of these boundaries is a place the same instrument can sharpen as more of the record becomes public.

What every verdict rests on. The arithmetic is exact; what it is exact about is a set of premises. In the order a ship travels through them, each with the rung its checking reached — validated (against the thing itself, held out), compared (against a model that keeps more), verified (the code against an exact solution of its own equations) or assumed (adopted, chosen or read; no machine here checked it):

1. assumed
   The registry's reports each ship's annual fuel, CO2eq and voyage split as its company reported them and its MRV verifier checked them; taken as reported. The CO2eq is the registry's, matched against the oil rows with the pinned GWPs; whether the registry used the same GWPs is not checked here, and the spike in §2b sits where no pinned row does.
2. assumed
   The rule's constants read and transcribed from the pinned Official Journal bytes and RED II Annex V; the battery re-derives the limits, a penalty and a flip threshold from them against hand computations from the same constants, so the transcription itself was read, not machine-checked.
3. assumed
   The fossil-oil hypothesis a ship whose reported CO2eq per gram of fuel some mix of the three pinned oil rows reproduces ([3,1688, 3,2609] g/g) burned such a mix. The screen prunes; it does not prove the fuel: any other fuel with the same ratio passes it.
4. assumed
   One fuel mix on every voyage the FuelEU scope share is the ship's CO2eq voyage split read as an energy split, which holds if the mix is the same on every voyage.
5. assumed
   2030, re-sailed the 2030 figures re-sail 2025's operations unchanged under the 2030 limit: a chosen scenario, not a forecast.

The arithmetic is exact; the weakest premise is assumed. Every verdict and every euro on this page is decided given these premises, and none of them is checked here against a ship.

For every non-compliant ship the instrument computes the exact energy fraction that must come from a zero-well-to-wake fuel for the same operations to comply in 2030 — an interval, because the fuel-type opacity propagates. That number is what a charterer, a lender under the Poseidon Principles, or a fuel supplier actually negotiates about; today it is bought as consultancy estimate. Here it is a computation anyone can rerun, and the same arithmetic generalizes to any candidate fuel intensity — replace zero with the fuel's pinned WtW value and the blend fraction is one exact division away.

This page is first light for the observatory, not its final shape: the roadmap is per-fuel candidate boxes (methanol, ammonia, wind-assist), company-level aggregation with pooling arithmetic, AIS-replay voyage scoping, and committed FORECASTS of next year's margins graded by the Forecast Gym's machinery. The instrument, the pinned rule, and the registry join are the load-bearing parts, and they are live above.