I turned a 78-criterion green software standard into a 20-point scorecard

EOF
France's national green software standard has 78 criteria. It is good. It is free. It is written by the telecom regulator, the media regulator and the environment agency together (RGESN v2, 17 May 2024). Every criterion comes with an implementation sheet.

And in two years I have never once seen it decide anything.

Not because it is wrong. Because a checklist and a decision are different objects, and almost every sustainability framework I have read is the first one pretending to be the second.

A maturity score is not a comparison

The RGESN does have a score. Here is what the regulator says it is for:

"Afin de faciliter le suivi en continu des efforts entrepris, il est aussi possible de calculer un score d'avancement renseignant son niveau de maturité vis-à-vis du référentiel."

A progress score. Your maturity, against the standard, over time. That is a genuinely useful thing and it is not the thing anyone actually needs at the moment of choosing.

Because the moment that matters is this one: two suppliers, two proposals, one budget, and someone has to say this one, not that one, in writing, in a way that survives a challenge. At that moment a 78-item checklist gives you nothing. Both bidders will tick most of it. Both will tick the cheap ones. Neither will tick the one that would have mattered.

The missing layer is not more criteria. It is weights and the honest admission that weighting is where the actual opinion lives.

Weighting forces you to say what you believe

I write these scorecards for digital procurement, so I had to pick. Three things pushed the weights around.

1. The footprint is not where the dashboard is. The French environment agency and telecom regulator measured it together and put it plainly: "79 % de l'empreinte carbone du numérique provient de nos équipements" and of that equipment footprint, manufacturing accounts for "80 % !" (ADEME–Arcep study, March 2023). Devices, and mostly the making of them. Server hosting is the part everyone can buy their way out of, and the small part. So the heaviest line in my scorecard is not "green hosting". It is: does this service still work on an old device, and can you prove it.

2. Page weight is a proxy, not a truth. I wrote about this separately — a 45 KB page can still cost a phone and the short version is that per-byte carbon models systematically miss the cost of making the device do work. So bytes get points, but fewer than the thing bytes are standing in for.

3. A criterion nobody can verify is worth zero points. This is the rule that changed the document most. Every line has to name the artifact that proves it, a build output, a Lighthouse run, a test matrix, a commit. "Committed to eco-design principles" is not a criterion. It is a sentence.

The scorecard

Twenty points, four to six lines. Here is the shape I converged on for a public-facing digital service.

Note what is not here. No certification points. No "has an environmental policy" points. No points for a carbon calculator screenshot. Every one of those is available to a supplier who changes nothing.

Note also the weighting choice that will annoy people: hosting is worth 2 out of 20. That is deliberate, and it is the direct consequence of where the footprint is. If a scorecard puts 8 points on green hosting, it is scoring the easy purchase, not the hard engineering.

Three rules that survived contact

• Weights before criteria. If you cannot say which two criteria you would trade against each other, you do not have a scorecard, you have a wish list. Write the weights first and let them force the list down to six lines.
• Name the artifact or drop the line. Every point has to be earnable by producing a file. If the only possible evidence is a claim, the line is greenwashing infrastructure.
• Score the thing the buyer cannot buy. Anything a supplier can acquire in an afternoon (a certificate, a host, a badge) belongs at the bottom. The points belong on the work.

If you are the one answering

The same document read backwards is a spec. If a buyer scores you on a minimum device profile, then the useful engineering work is a test matrix and a throttled run, not a sustainability appendix. If they score trajectory, then one dated measurement now is worth more than any prose later. And if they score hosting at 2 points, the greenest region in the world will not save a 4 MB bundle.

Most of the sustainability work I see suppliers do is aimed at the lines that are cheap to claim, because those are the lines most scorecards weight highest. That is a scorecard problem before it is a supplier problem.

The tool, and its limits

I built a generator that produces this — the criterion, the execution condition, the 20-point scale and the documents to require from six questions: perennova-eco.github.io. It runs entirely in your browser, no account, nothing sent anywhere, free.

Two honest limits. It is written for French public procurement and the legal hook is Article 35 of the 2021 Climate and Resilience Act, whose implementing decree (décret n° 2022-767 of 2 May 2022) removes "la faculté de sélectionner les offres sur la base du critère unique du prix" and applies "au plus tard […] en août 2026" (DAJ, Ministry of the Economy) — and it outputs French. If you write tenders elsewhere, the part that transfers is the table above and the three rules, not the generated text.

If you have built one of these and weighted it differently, I would like to know which line you moved and why. The weights are the argument; everything else is bookkeeping.

(Previously: your servers are not the problem — devices are 65–92% of digital's footprint.)