Methodology
Version 0.1.0. The method is public, versioned and dated (verifiable anteriority); the code and the grid are on GitHub.
The grade in one sentence
Each project receives two grades from A to E — one for the site (the context the project endures) and one for the project & process (what the operator chooses) — coupled with a documentation confidence level. A grade never appears without it; below 40% coverage we write “insufficient data”. We do not grade the unknown.
The score is computed from the sourced values of each project's file, and nothing else.
🤓 Reproducible method: anyone who clones the repository and runs make score lands exactly on the displayed letters.
The scale
An A is proven, never promised
The scale has a precise meaning, locked by construction: E — opaque, or verified unfavorable · D / C — partial or good promises · B — complete, quantified promises and an exemplary process · A — reserved for verified operation. A fully declarative file, however excellent, mathematically caps below the A: announced values are graded with caution (“announced ≠ measured”), and only proof lifts that cap. Nobody “receives” an A — it is demonstrated.
Operators — how to earn your A
Bring third-party proof of your values in operation: a copy of your EED filing to the regulator, an ISO/IEC 30134 audit certificate from an accredited body, or an inspection-office attestation — never a marketing page. Submit it through the public correction channel (source required, declaration of interest). The data then flips from “announced” to “measured”, the declarative cap lifts, and the re-score is traced in the public audit log. No letter to wait for, no payment — the proof does everything.
Two clarifications that protect everyone. Proof lifts your project & process grade — never the site grade: the territory (water, grid, land) is endured, it cannot be proven away. And no capture is possible: no payment, third-party proof, a grade that moves only because verified facts moved.
The 5 pillars
- Energy25 %
- Water20 %
- Land & biodiversity20 %
- Local impact20 %
- Transparency & governance15 %
The grade, like a school report
A data center receives a report card. The five main themes — the pillars — are the subjects: Energy, Water, Land & biodiversity, Local impact, Transparency & governance. In each subject, several “tests” — the indicators — produce marks.
Stage 1: within each pillar, indicators are averaged by their internal coefficient (one test counts more than another).
Stage 2: pillar scores are averaged by the weights above. One principle: no crossover — an indicator counts in its pillar only, never elsewhere.
Example (2 pillars, scores already normalised to 100)
- Energy (weight 25%) — carbon 80 (coef 0.6) · PUE 40 (coef 0.4) → 80×0.6 + 40×0.4 = 64
- Water (weight 20%) — stress 30 (coef 0.7) · consumption 50 (coef 0.3) → 30×0.7 + 50×0.3 = 36
- Global (weights renormalised 25/45 and 20/45) → 64×0.56 + 36×0.44 ≈ 52 → letter C (threshold 50–64)
Two grades — and it is not the average of the pillars
This is the nuance that surprises. The two headline grades do not aggregate by pillar but by block: the site grade = the endured indicators (what the territory imposes — grid carbon, water stress, land); the project & process grade = the chosen indicators (efficiency, cooling, consultation, transparency). The pillar sub-scores group each theme, mixing endured and chosen. The two groupings are orthogonal: a grade does not read as the average of the displayed pillars.
From measurement to grade — two families of thresholds
Each indicator turns a raw value into a 0–100 sub-score through a threshold. That threshold is set in two ways, and the distinction is declared, not hidden:
- Family A — external referential. We anchor on an existing, sourced, defensible standard (ISO 30134 for efficiency, the Water Framework Directive for water-body status, WRI Aqueduct for water stress…).
- Family B — declared editorial ratio. Where no norm exists (“what size is too big?”), we do not invent an absolute threshold: we express a contextual ratio (project relative to the municipality, to the basin) and own the editorial choice.
Rule: maximise the referential, minimise the editorial — and declare it. The “Threshold referential” column in the detail below states, indicator by indicator, what each threshold rests on.
Locked by construction
- Announced ≠ measured: a figure declared by the operator is tagged “announced”, graded with caution, re-verifiable ex-post.
- Transparency floor: disclosing a data point can never degrade the grade relative to hiding it.
- Ethical lock: a municipality's vulnerability can never improve a project's grade.
- “Project ≠ grade” firewall: we help the project, never the grade. The public grade moves only when the real project changes (sourced facts, audit-logged re-scoring) — never because an operator paid.
- Governance: every grid evolution goes through a tagged version and a justified, signed changelog entry.
Status, limits & calibration
This version is preliminary and may evolve. What the grade is not: a statistically validated predictive model, nor an oracle. What it is: an expert, sourced, versioned and reproducible grid, applying thresholds anchored on public referentials and declaring its editorial choices.
- Provisional weights and thresholds: they are calibrated by retrospective validation on known-outcome cases (accepted / abandoned projects), not asserted a priori.
- Traceability: every grade is dated and tied to a methodology version; an “A” means “the best achievable today”, not “the best forever”.
References & standards
- ISO/IEC 30134 (2 PUE · 3 REF · 4 CUE) — data centre performance indicators
- EU Code of Conduct for Data Centres (JRC)
- Energy Efficiency Directive (EED) 2023/1791 — data centre reporting
- WRI Aqueduct — basin water stress
- Water Framework Directive 2000/60/EC — water-body status
- ADEME Base Empreinte — grid emission factors
- RTE eCO2mix / Caparéseau — grid carbon and capacity
- “Social Licence to Operate” framework — social acceptability of projects
See the 24 indicators in detail
Each indicator carries its coefficient (weight within the pillar), its scoring direction, the basis of its threshold (external referential or declared editorial choice) and its data source — a schema constraint, not a best practice.
Energy25 %
| Indicator | Block | Coef. | Direction | Threshold referential | Source |
|---|---|---|---|---|---|
| Local grid carbon intensityHow much CO₂ is emitted to produce the electricity the site will consume, based on the regional grid. | Site (endured) | 15 % | lower = better | ADEME Base Empreinte — grid emission factors | RTE eCO2mix |
| Grid connection capacity / proximityHow easily the site can be connected to the power grid: available capacity and distance to the nearest substation. | Site (endured) | 35 % | qualitative | — | RTE Caparéseau, S3REnR |
| Grid congestion / connection queueHow congested the local grid is: the more saturated, the more the connection competes with other users. | Site (endured) | 25 % | qualitative | — | RTE / ODRÉ |
| PUE (announced energy efficiency)The announced energy efficiency: total electricity used versus what actually powers computing. Closer to 1 is better. | Project (chosen) | 15 % | lower = better | — | ICPE filing, press |
Water20 %
| Indicator | Block | Coef. | Direction | Threshold referential | Source |
|---|---|---|---|---|---|
| Local water stressWater tension in the area: in an already water-scarce region, withdrawals weigh more heavily. | Site (endured) | 30 % | qualitative | ZRE zoning + Propluvia restriction levels (French regulatory framework) | ZRE, Propluvia, Hub'Eau |
| Water body statusThe ecological health of the affected river or aquifer, per the EU classification (from 'high' to 'bad'). | Site (endured) | 20 % | qualitative | EU Water Framework Directive ecological status classes | Hub'Eau |
| Basin withdrawal pressureHow intensively water is already withdrawn from the basin: a heavily-tapped basin ill affords a new user. | Site (endured) | 20 % | qualitative | — | BNPE, Hub'Eau |
| Cooling technology + announced water volumeThe cooling technology chosen and the water volume it will consume, as announced by the operator. | Project (chosen) | 20 % | qualitative | — | ICPE filing |
Land & biodiversity20 %
| Indicator | Block | Coef. | Direction | Threshold referential | Source |
|---|---|---|---|---|---|
| Overlap / distance to protected areasHow close the project is to protected natural areas (Natura 2000, nature reserves…). | Site (endured) | 20 % | qualitative | — | INPN, Natura 2000, ZNIEFF |
| Soil status (artificialized / natural / agricultural)What the land was before: already built (brownfield), natural, or farmland — sealing farmland or nature weighs most. | Site (endured) | 35 % | qualitative | Cerema artificialization inventory + ZAN framework categories | Cerema, RPG |
| Ecological corridors (TVB)Whether the project severs a wildlife corridor (green-and-blue network). | Site (endured) | 20 % | qualitative | — | SRADDET, INPN |
| Project ground footprintThe project's ground footprint in hectares: the larger, the greater the land impact. | Project (chosen) | 15 % | lower = better | — | ICPE filing, building permit |
| Avoid-reduce-offset measures / heat recoveryMeasures to avoid, reduce or offset harm, and to reuse the heat produced. | Project (chosen) | 10 % | qualitative | ISO/IEC 30134-6 — Energy Reuse Factor (ERF), le référentiel international de la réutilisation de la chaleur fatale des data centers | operator, permitting file |
Local impact20 %
| Indicator | Block | Coef. | Direction | Threshold referential | Source |
|---|---|---|---|---|---|
| Municipality socio-economic profileThe host town's social profile. A safeguard: a project must never improve its grade by locating in a vulnerable town. | Site (endured) | 15 % | nuanced | — | INSEE |
| Absorption capacity (project size vs municipality)The project's size relative to the town: a very large project in a small town disrupts more. | Site (endured) | 15 % | lower = better | — | INSEE |
| Technological hazard / nearby Seveso sitesThe presence of hazardous industrial sites (Seveso) nearby, adding risk to the surroundings. | Site (endured) | 10 % | qualitative | — | Géorisques |
| Announced jobsThe number of permanent jobs announced, relative to project size. More is better. | Project (chosen) | 15 % | higher = better | — | press releases, press |
| Announced tax revenueThe tax revenue the project would bring to the town, as announced. | Project (chosen) | 10 % | qualitative | — | council deliberations, press |
| Observed level of oppositionThe scale of observed opposition: appeals, petitions, mobilizations reported by the press. | Site (endured) | 20 % | qualitative | — | press, petitions, local groups |
| Elected officials' position (voted facts only)Elected officials' position as expressed by official votes only (council decisions), with no personal judgment. | Site (endured) | 15 % | qualitative | — | council deliberations, recorded votes |
Transparency & governance15 %
| Indicator | Block | Coef. | Direction | Threshold referential | Source |
|---|---|---|---|---|---|
| Consultation / consent / governance (procedural proxies)The quality of the democratic process: public inquiry held, environmental authority opinion, appeals, CNDP referral. | Process | 60 % | qualitative | — | public inquiry, environmental authority opinion, CNDP, legal appeals, council deliberations |
| Documentation transparency / file availabilityHow genuinely accessible the project's file is to the public: published studies, figures and commitments. | Process | 40 % | qualitative | — | registries, official publication |