Methodology & curation policy

How the UK energy events corpus is selected, sourced, verified and corrected. Published because a dataset you cannot audit is a dataset you should not cite.

This site renders 54 GB-relevant events. The CC-BY-4.0 export in data/events.csv carries 56 rows — it also includes non-GB events (currently two German ones) that are excluded from GB-scoped views and feeds.

The editorial premise

Every source has incentives — including the ones we agree with. Government bodies serve the current administration. NGOs need alarming numbers to justify their existence and fund their work. Industry bodies exist to advocate for members. Academics face publication bias toward novel findings. None of this makes any of them liars; it makes their framing unreliable while their measurements may be excellent.

The working rule: use sources for their raw data, not their interpretations or policy recommendations. The same dataset tells opposite stories depending on who presents it.

Seven-tier source hierarchy

Sources are ranked by distance from the measurement, not by institutional prestige.

  1. Tier 1 Raw measurements

    Instrument and settlement data — Elexon BMRS, NESO operational data, Sheffield Solar PV_Live, National Grid Carbon Intensity. Numbers before anyone has interpreted them.

  2. Tier 2 Peer-reviewed research

    Journal articles with declared methods, sample sizes and funding. Check funding declarations: publication bias favours novel and alarming findings.

  3. Tier 3 Government statistical releases

    DESNZ Energy Trends, Defra agricultural census, ONS, JNCC indicators. The underlying data is generally sound.

  4. Tier 4 Government reports and framing

    The same bodies, writing narrative. Serves the administration of the day. Use the tables, not the conclusions.

  5. Tier 5 NGO-collected data

    Survey and monitoring data from conservation bodies. Collection may be rigorous; framing is always in service of fundraising.

  6. Tier 6 Industry data

    Trade-body statistics. Exists to advocate for members — will minimise sector costs and maximise sector benefits.

  7. Tier 7 Campaign material

    Press releases and advocacy output from any side. Cited only to characterise a position, never as evidence for a fact.

The incentive check is mechanically enforced

Every row carries a methodology field recording verification status, source-tier classification, and — where any non-primary source is used — an explicit incentive note. This is not an honour system. The validator at scripts/events/validate-seed.ts rejects at commit time any row citing an NGO or industry-advocacy host (RSPB, WWF, Energy UK, RenewableUK and similar) without that note. A second gate validates the export against a Frictionless Data table schema.

What qualifies as an event

An event must have a defined occurrence window, at least one primary-source URL, and a demonstrable connection to the GB electricity system. Announcements are recorded as announcements — a final investment decision or Royal Assent is a paper event, and the corpus does not imply it produced a measurable change in generation at that date. This distinction matters for anyone doing before/after analysis: operational events (storms, plant closures, interconnector outages) may show short-run signal; policy events generally will not, and any short-window delta around them is confounded by weather, demand seasonality, fuel prices and overlapping events.

Categories (taxonomy v1, frozen 2026-04-25)

  • policy — Ofgem decisions, price caps, subsidy changes, RIIO rulings
  • infrastructure — Plant retirements, interconnector events, FIDs, transmission build
  • geopolitical — Invasions, sanctions, OPEC decisions, pipeline incidents
  • market — Gas price extremes, capacity auction outcomes, day-ahead records
  • weather — Named storms, dunkelflaute, heatwaves, cold snaps, wind droughts
  • demand — Lockdowns, industrial closures, structural demand shifts
  • milestone — Coal closures, generation-share records, demand records
  • anomaly — Algorithmically detected deviations (Sprint 3+) — none published yet
  • counterfactual — Hypothetical scenarios (Year 2+) — table not yet created

Severity

minor (footnote-worthy) · notable (default annotation) · major (always labelled) · pivotal (capstone event, shaded band). Severity is an editorial judgement about significance to the GB system, and we say so rather than dressing it up as a computed score.

Status and retractions

Rows are draft (not public), published, or retracted. Every change is captured in an append-only revisions table, and each retraction records a reason.

Current limitation, stated plainly: public surfaces load through an RPC that returns only published rows, so a retracted event is removed from this site and from both feeds rather than shown struck through. The reason is retained internally and in the revisions table, but is not yet rendered publicly. Surfacing retractions with their reasons is outstanding work; until it lands, treat disappearance from the corpus as the retraction signal and consult the repository history for the reason.

Known limits

  • Coverage is GB-centred. Non-GB events appear only where they moved the GB system.
  • Selection is editorial. There is no algorithm deciding what matters; a human did, and that human has a stated position.
  • Underlying grid data is national aggregate by fuel type — there is no per-plant or per-wind-farm resolution, so plant-level attribution is not possible from it.
  • Pre-2017 events predate the daily rollup, so no grid mini-chart is shown for them.

Corrections

If a row cites a non-primary source where a primary source exists, or a figure is wrong or stale, open an issue with the primary URL and we will substitute it and update the methodology note. Corrections are made in the open. A live URL is not the same as a current figure — check the vintage of the data, not just that the page loads.

Citing this corpus

The corpus is CC-BY-4.0 and archived on Zenodo under concept DOI 10.5281/zenodo.20106845. Subscribe to dataset changes via the events feed — separate from the main feed, so you can follow the data without the opinion. Browse the corpus at UK Energy Insights.