Methodology changes

From version 2.2 to 3 — 10 November 2026

Why we changed. In version 2.2 about half of the differences between scores came from how many recent documents were public, and the final number was the judgement of a language model. A well-documented project could outrank a more solid one. We want the number to be about the credit, and we want anyone to be able to redo the calculation.

What is better.

• The number is a «document-based rating»: how solid the credit is, as far as the registry's public documents show.
• Four dimensions: additionality, quantification, permanence (only where carbon is stored), claim safety.
• Transparency and documentation no longer enter the number. They become the «evidence coverage» (high, medium, low), shown beside it.
• The number comes from a public table: a starting point for the type of project, taken from public sources (the ICVCM's decisions on the methodology and CCQI's scores by project type), then additions and deductions on the project's own facts. The model reads the facts; it no longer gives the score.
• Each rule has its source, downloaded and quoted.
• Where the evidence is not enough, the page does not give a low number: it says that it is too early to rate the project, or that the project is not yet assessable.

What changes in scores (on 6,345 pages with a 2.2 score, in the trial recalculation):

• 4,779 pages receive the version 3 rating. The median goes from 6.2 to 4.6. One page rises by more than one point (by 1.3). 2,399 fall by more than two.
• The scale is lower on purpose: 5.0 means «no public source says good or bad». Four bands: high (6.5 or more, 222 pages), moderate (5.0–6.4: 1,725), low (3.5–4.9: 1,386), very low (under 3.5: 1,446).
• The types the public sources judge weaker fall most: avoided deforestation under the earlier methodologies, improved forest management, cookstoves, grid-connected renewable electricity. Durable removals, biochar, methane and landfill gas under a methodology the ICVCM approved stay higher.
• 1,092 pages lose the number: no validation or verification report was read. They show «Too early to rate» (578) or «Not yet assessable» (514), with the registry's data.
• 474 pages keep their 2.2 score with its label, because a fact the new table needs was not collected when they were analysed.
• The fNRB rule keeps its sizes and now counts on quantification: it is a risk of counting too many credits.
• Six v2.2 sector rules are kept with their v2.2 sizes, each on its v3 dimension. Each applies only where the documents read do not state the fact.

What stays the same. The sources: only the registries' public documents. Each figure with its quote. Sellers' documents outside the score. The absence of a score says nothing about quality. A published score changes only to correct an error; new criteria need a new version, as this one. The version 2.2 page stays online, unchanged, at a fixed address, and pages with a 2.2 score link to it.

The rules of this version

From version 2.1 to 2.2 — October 2026 (2.1 scores withdrawn on 8 November 2026)

Why we changed. In version 2.1 a project was assessed with general rules for its sector. We wanted each project read against the official text of its own methodology, and each figure on a page shown with its quote in the document it comes from.

What is better.

• Each project is assessed by the rules of its own methodology.
• A figure is kept only when the sentence that states it was found on the page of the document.
• The page's texts are built from the facts, with fixed sentences. The limits on a score are applied by code, the same for every project.
• Where no rule covers a project's methodology, the page shows registry data only, with no score.

What changes in scores. The analyses made with version 2.1 came before the checks the method applies today: the sector rules, the fNRB rules and the others that came in from 1 November 2026. A reading showed that an analysis made that way could carry statements the documents do not support. So on 8 November 2026 we withdrew every 2.1 score, and did not leave online scores made under rules we no longer apply: on that date about 8,400 pages show registry data and «Re-assessment pending». Each project is reassessed with the current method. Every score online is a version 2.2 score.

What stays the same. The four dimensions and their weights. The sources: only the registries' public documents. Documents found only at a seller never enter a score. The absence of a score says nothing about a project's quality.

The rules of version 2.2, as they were