Transparency

How we read the data

Substrato mixes open inventories and curated corridors. This page explains what each reliability label means, who verifies the links, and which limits stay explicit.

Revised:

We use two distinct levels: confidence on the point (how reliable the site's existence and properties are) and certainty on the link (how well documented the relationship between two sites is). They are not the same thing. An OSM data center can have low confidence and no edges; a named PPA can have confirmed certainty even if a nearby point stays raw.

Two layers on the map

Basemap (inventory)

Points bulk-imported for geographic context: energy plants from the WRI Global Power Plant Database (Italy + EU ≥ 100 MW) and data centers from OpenStreetMap (telecom=data_center tag in the EU area). OSM for data centers is incomplete and uneven: many campuses are missing, some are misplaced, quality varies by country. WRI is more structured, but remains a global database with delays and import thresholds. Imported records do not claim to be a supply chain.

Curated corridors (project thesis)

A few sites enriched by hand (descriptions, impact, labor, sources) and connections with mandatory certainty + evidence_note. This is where the line «the cloud has a body» lives: documented energy, AI silicon, battery metals, labor contrast. We prefer fewer true arrows to many suggestive ones.

Confidence on the site

Measures how reliable the point record is (identity, location, declared capacity).

High
Official document or primary operator statement (corporate fact sheet, annual report, PPA named on the site).
Medium
Journalistic source or coherent secondary report; typical of imported WRI plants and many curated records still being enriched.
Low
Indirect clue. Default for data centers imported from OpenStreetMap until there is secondary verification.
Estimate
Value modeled or estimated by the dataset author (e.g. average impact factors), always labeled as such on the card.

Certainty on the link

Measures how well documented the relationship between two sites is (powers, supplies, manufactures_for). Every new edge requires certainty and an evidence note in Italian.

Confirmed
Direct proof of the link: named PPA or supply contract, joint statement, corporate report that explicitly cites origin and destination.
Likely
Strong geographic or market evidence without site-to-site proof (same basin and same supplier of a known agreement; recognized ODM of a hyperscaler without the campus packing list).
Inferred
Transparent working hypothesis: grid proximity, same product class, energy cluster. Never presented as fact. Example: Turbigo → Milan campuses without a public PPA.
Disputed
Conflicting sources or relevant public contestation of the link or project. Stays on the map with the label; it is not «resolved» quietly.

How a grade is raised

The grade rises only with new evidence, not for narrative convenience.

  • From inferred to likely: coherent evidence is required (same operator/basin contract, documented market supply chain) and a note explaining why it is not yet confirmed.
  • From likely to confirmed: a named primary document is required (PPA, joint statement, report naming both parties). Without that the link stays likely or drops.
  • powers without a named PPA or offtake: never confirmed.
  • supplies / manufactures_for only if the material or product belongs to that supply chain (no lithium→chip or cobalt→logic foundry).
  • A site without an edge is valid: projects, context, inventories. We do not invent links to fill the map.

Who verifies

Editorial curation is by Relatronica (Substrato project): corridor selection, certainty/confidence compilation, evidence notes, and source updates. OpenStreetMap and WRI remain under their communities and licenses; we do not «certify» the entire automatic inventory. Community corrections go through Report (preferred: traceable GitHub issue) and are incorporated after human review — there is no anonymous direct write access to the published dataset.

When it is updated

Each curated record may have updated_at; the dataset_meta.json file reports the date of the last merge. We do not offer real-time updates. Typical cadence: when relevant PPAs or reports appear, when a report is verified, or in corridor enrichment batches. When citing, always note the date you consulted the map.

Main sources and limits

WRI Global Power Plant Database

Energy plants, CC BY 4.0 license. Import: all plants in Italy + EU ≥ 100 MW. More rigorous than OSM on structure and capacity, but not an offtake toward data centers.

OpenStreetMap (Overpass)

Data centers via telecom=data_center tag, ODbL license. Low confidence by default. Incomplete and uneven: useful as a basemap, insufficient alone for journalism on a single site without verification.

Curated records and connections

GeoJSON in data/curated/ and edges in connections_edges.json: descriptions, labor, impact, certainty. Always take priority over automatic candidates in case of conflict.

Context overlays

Power grid (OpenInfraMap), water stress (Aqueduct × HydroBASINS), submarine cables (historical extract). Off by default; they are not supply-chain links.

What we do not claim

  • Completeness of the European inventory of data centers or mines.
  • Physical electron-by-electron or wafer-by-rack traceability without a document.
  • Invented employment numbers: we prefer null + a note to a cosmetic 0.
  • Institutional certification or third-party audit of the dataset.

How to cite

Substrato (Relatronica), map consulted on YYYY-MM-DD, substrato.eu — also note the link certainty if you use a specific connection, and the primary source linked on the card when present.

Open the map Report an error Press contacts

Updated: 2026-09-27. For corrections on a link or source use Report or open a GitHub issue.