RemitScan is a deterministic surveillance index. Every flag in this product can be reproduced from raw public data using the rules and parameters described below. There is no machine-learning inference in our flag generation, no proprietary asset list, and no opinion. This page is the spec.
RemitScan ingests from five primary source classes. Each ingest is timestamped to nanosecond precision at receipt, and the raw payload is retained in an append-only object store with SHA-256 content addressing.
We maintain a canonical asset registry keyed by Energy Identification Code (EIC). 4,926 generation assets across 27 countries are mapped to (a) market participant, (b) primary fuel, (c) bidding zone, (d) TSO, and (e) WGS-84 coordinates with ±1km tolerance.
Where an MP submits an unmapped EIC, the registry is updated within one business day. Mappings are deterministic — no fuzzy matching is used at the registry layer.
The matching engine pairs UMM disclosures with TP unavailability records. The pairing function is:
MATCH(umm, tp_event) ⇔
asset_eic(umm) = asset_eic(tp_event) [hard]
∧ |start(umm) − start(tp_event)| ≤ Δt [Δt = 48h]
∧ Jaccard(cause_terms(umm), cause_terms(tp_event)) ≥ 0.4
∧ |mw_declared(umm) − mw(tp_event)| / mw(tp_event) ≤ 0.20EIC equality is required. Time tolerance Δt is parameterised per regulator interpretation; the default is 48h to account for late filings within the "without delay" regime. Cause-term Jaccard guards against spurious co-occurrence; the 0.4 threshold was calibrated against a 2024 supervised set of n=12,400 hand-paired records and yields F1 = 0.94.
Each unavailability event is scored 0–100 across four components, additive with caps:
Day-ahead price impact component (25 pts) is in development and will be added once the DA price feed is live.
Materiality determines surveillance prioritisation only — it never affects whether a flag is raised, only its display ranking.
Three deterministic rules are live in v1 (DE+FR active tier; watch-tier statistics for the remaining 34 ENTSO-E countries). Two further rules are under development and are not yet emitting flags.
The DQI is a 0–100 composite per MP, computed as a weighted average over a trailing 90-day window:
DQI is published per MP, per country, and as a sector aggregate. It is recomputed nightly and is fully reproducible from inputs.
Every flag carries a lineage manifest enumerating: raw record SHA-256, query manifest SHA-256, calculation step SHA-256, rule version, and detection wall-clock timestamp. Manifests are signed by the publishing pipeline and verifiable offline.
Any reader of a RemitScan evidence pack can replay a flag from raw data using the public AL-RUNNER tool and verify byte-for-byte equivalence.
RemitScan does not assert market manipulation. Flags are statistical disclosure-quality indicators; their resolution requires the judgement of operators and regulators with access to non-public information. Sub-100 MW events, behind-the-meter generation, and aggregator-managed virtual assets are out of scope for v2.5.
As of the v2.5 effective date, three known gaps in the ingested source data prevent us from emitting flags we would otherwise be able to emit deterministically. We surface them here rather than in a footnote because they materially affect what the inbox shows today.
These three items are the dominant reason a regulator looking at the live inbox today sees fewer flags than the rules would otherwise produce. They are not silenced or filtered — they are not detectable from current Gold-layer state. Each is tracked, dated, and will be reflected in the changelog when resolved.
This document is reviewed quarterly by the Methodology Board.