Methodology 1.2
How the Veridio Score works
One 0–100 number per project, fully itemised: 8 independent factors, each with its own evidence, combined by fixed weights. The numbers on this page are read live from the scoring code, so they cannot drift from what we actually run.
Factors, weights and where the data comes from
Scroll sideways to see every column →
| Factor | Land use | Other types | Data source |
|---|---|---|---|
| Methodology quality | 15% | 20% | CCQI pillar scores (the weakest pillar counts double) and the ICVCM Core Carbon Principles decision on the methodology. |
| Project documents | 20% | 20% | Veridio's deterministic 0-100 document score over the project's own PDFs (design documents, monitoring, validation and verification reports), adjusted by the due-diligence read: -25 when the documents undermine the claims, +5 when they support them. |
| Certification & standard | 6% | 15% | ICVCM programme status, CORSIA eligibility and the project's registry status. |
| Country governance | 10% | 10% | World Bank Worldwide Governance Indicators: rule of law, control of corruption, government effectiveness, political stability. |
| Delivery track record | 4% | 10% | Registry issuance and retirement records: volume issued, share retired, how recently the project issued. |
| Over-crediting evidence | 20% | 10% | Published project-level achievement ratios (share of claimed reductions found real) and documented controversies; without a study, Veridio's biomass cross-check of credits issued per hectare. |
| Satellite record land-use projects only |
18% | n/a | Veridio's remote-sensing verdict for the site (land-use projects only). Verdicts from a map pin rather than a real boundary are damped. |
| Natural hazards land-use projects only |
7% | n/a | Measured exposure of the site to fire, drought (public drought index), flood, cyclones and deforestation pressure around it (land-use projects only). The worst hazard sets the factor score. |
| Total | 100% | 85% → renormalised | For non-land projects the satellite and hazard weights do not apply; the rest is rescaled to 100. |
Each factor score maps to a five-step risk scale, the inverse of the score: 85+ very low · 70+ low · 50+ medium · 30+ high · below 30 very high risk.
Missing data, coverage and confidence
Missing data is never scored as zero or as average. A factor without data is left out and the remaining weights are renormalised. Coverage is the share of the applicable weight that is backed by real data — for a cookstove or wind project the satellite and hazard factors do not apply and are not counted against it.
Scroll sideways to see every column →
| Coverage | Confidence | What we publish | Projects today |
|---|---|---|---|
| 75% or more | High | Headline score | 78 |
| 50–75% | Medium | Headline score | 2,865 |
| 40–50% | Low | Headline score, flagged | 3,164 |
| below 40% | Insufficient | No headline score; factors still shown | 7,153 |
Hard rule: a project that fails one of our reject rules (grade REJECTED) is capped at 35, whatever its factors say. The A–AAA grade remains the sell / don't-sell gate; the score is the continuous, explainable layer beside it.
Document score
We download each project's own PDFs — design documents, monitoring and annual reports, validation and verification reports, non-permanence risk reports — and score them deterministically on five components of 20 points each: completeness, verification, quantification, permanence and integrity. Every point is a text match stored with its snippet, so each score can be traced to a quote.
Limits. Text analysis measures what is documented, not whether it is true. Document discovery currently covers Verra and Acorn only.
Satellite record
Every site page says which rung of the boundary ladder its geometry stands on: 1 the project's own boundary file from its registry (for example Gold Standard GIS files), 2 the open project-boundary database (Karnik et al. 2024), 3 coordinates listed in the project's documents (an approximation), or 4 the registry's map pin plus a 5–15 km window sized from the reported hectares. Rungs 1–2 are real boundaries (medium confidence); 3–4 are approximations (low confidence). The control is a ring from 1.3× to 2.0× the site radius. For each sensor we take site minus control per year and compare the mean before and after the project start (difference-in-differences). A signal only counts when it exceeds the larger of a fixed floor and twice the pre-project noise.
Scroll sideways to see every column →
| Sensor | Product | Use |
|---|---|---|
| MODIS | MOD13Q1 NDVI, 250 m, 4 composites a year, 2012 onwards | Long pre-project baseline |
| Sentinel-2 | L2A 10 m, 2 clearest scenes a year, 2017 onwards | Sharper, shorter baseline |
| Land cover | Impact Observatory annual 10 m, 2017–2023 | Tree-cover %, corroborates only |
Verdicts: ▲ supports and ▼ contradicts need a vegetation signal; land cover alone never decides. ● neutral means nothing distinguishable from the surroundings. All imagery comes from Microsoft Planetary Computer.
Soil-carbon and grazing claims get no satellite verdict: greenness from orbit cannot test them, so for those projects the factor stays empty rather than guessing.
Limits. Pin sites are always low confidence: a landscape window dilutes scattered smallholder plots, and young agroforestry needs 3–5 years to show at these resolutions — neutral on a young pin site is the expected outcome, not a red flag. Low-confidence verdicts are damped to 35% of their effect on the score. The start year defaults to 2021 (flagged “assumed”) until confirmed. The control ring may itself contain project plots, which biases toward neutral.
Natural hazards
Five hazards, all measured from open data, none modelled. The worst one sets the factor score: a fire-prone site is not made safe by being flood-proof.
| Hazard | Measure | Source |
|---|---|---|
| Fire | share of the site burned per year, last five full years | NASA MODIS MCD64A1 burned area, 500 m |
| Drought | share of the last ten years' months in severe or extreme drought (Palmer Drought Severity Index ≤ −3), compared with the decade before; levels: under 1 / 1–2 / 2–3.5 / 3.5–5 / 5+ years of the decade. The site's own vegetation record corroborates | TerraClimate PDSI, ~4 km, monthly (Abatzoglou et al. 2018) |
| Flood | share of the site that was open water sometimes but not permanently, 1984–2020 | JRC Global Surface Water, 30 m |
| Cyclone | hurricane-strength storms passing within 100 km, per decade, last 30 seasons | NOAA IBTrACS best tracks |
| Deforestation pressure | tree cover lost per year in the landscape around the site | Impact Observatory land cover, 10 m |
Limits. Past exposure is not a forecast. Drought is read at the site centre on a ~4 km grid; PDSI rises with warming-driven evaporative demand, so recent decades read drier than the past in many regions. Hazards on an approximate site describe the area, not the plots.
How the weights are set — and how much they matter
The score aggregates evidence; it is not fitted to an outcome and does not predict over-crediting (see the backtest below). Weights are set in advance by a rubric: independent, project-specific evidence weighs most; developer-authored or type-level evidence less. They change only with a new methodology version, by declared rules: towards external ratings only once 25 projects are paired (today: 1), and away from any factor that ranks documented over-crediting cases above their peers. They are never searched to maximise separation.
| Factor | Independent of the developer? | About this project? | Why this weight |
|---|---|---|---|
| Methodology | yes — CCQI, ICVCM | no — project type | a sound method is necessary but says nothing about execution |
| Documents | no — the developer's own PDFs, scored by us | yes | the only project-specific evidence for most projects; shows what is documented, not truth |
| Certification | partly — registry and programme record | partly | on land projects it ranked documented over-crediters above their peers, so it weighs less there |
| Country | yes — World Bank | no — national | governance context for permanence and enforcement |
| Delivery | registry record | yes | measures execution, not integrity; pointed the wrong way on land projects |
| Over-crediting | yes — peer-reviewed studies, biomass cross-check | yes | the strongest evidence where it exists |
| Satellite | yes — measured | yes | the only independent measurement of a land claim |
| Hazards | yes — measured | yes — the site | exposure to future reversal, not past integrity |
Sensitivity. Every rated project is re-scored under each alternative below with the live formula; the published set is held fixed. Spearman = rank agreement with today's weights; band flips = share of projects whose score crosses one of the 85 / 70 / 50 / 30 lines.
Scroll sideways to see every column →
| Scenario | Rank agreement | Median shift | 90th pct shift | Band flips | Top 10% kept |
|---|---|---|---|---|---|
| v1.0 weights | 0.998 | 0.0 | 0.0 | 0.4% | 99.7% |
| equal weights | 0.989 | 0.4 | 5.3 | 10.8% (above the flag threshold)▲ | 85.2% |
| no self-reported factors | 0.928 | 0.9 | 9.5 | 18.0% (above the flag threshold)▲ | 61.9% |
| methodology ×0.5 | 0.999 | 0.0 | 0.0 | 0.6% | 91.6% |
| methodology ×1.5 | 0.999 | 0.0 | 0.0 | 1.0% | 94.7% |
| without methodology | 0.996 | 0.0 | 0.0 | 1.5% | 89.0% |
| documents ×0.5 | 1.000 | 0.0 | 0.0 | 0.0% | 99.4% |
| documents ×1.5 | 1.000 | 0.0 | 0.0 | 0.1% | 99.7% |
| without documents | 0.999 | 0.0 | 0.0 | 0.2% | 98.4% |
| certification ×0.5 | 0.968 | 0.9 | 8.4 | 16.0% (above the flag threshold)▲ | 72.0% |
| certification ×1.5 | 0.983 | 0.4 | 5.5 | 9.8% | 84.7% |
| without certification | 0.819 | 3.6 | 23.1 | 43.3% (above the flag threshold)▲ | 33.0% |
| country ×0.5 | 0.990 | 0.5 | 4.4 | 14.4% (above the flag threshold)▲ | 86.5% |
| country ×1.5 | 0.995 | 0.4 | 3.3 | 3.8% | 89.6% |
| without country | 0.966 | 1.1 | 10.6 | 25.6% (above the flag threshold)▲ | 72.0% |
| delivery ×0.5 | 0.982 | 0.4 | 4.0 | 6.1% | 83.0% |
| delivery ×1.5 | 0.994 | 0.3 | 3.0 | 8.7% | 93.0% |
| without delivery | 0.929 | 0.8 | 9.5 | 18.1% (above the flag threshold)▲ | 62.1% |
| over-crediting ×0.5 | 1.000 | 0.0 | 0.0 | 0.0% | 100.0% |
| over-crediting ×1.5 | 1.000 | 0.0 | 0.0 | 0.1% | 99.8% |
| without over-crediting | 0.999 | 0.0 | 0.0 | 0.1% | 100.0% |
| satellite ×0.5 | 0.999 | 0.0 | 0.0 | 0.2% | 99.7% |
| satellite ×1.5 | 0.999 | 0.0 | 0.0 | 0.2% | 99.7% |
| without satellite | 0.997 | 0.0 | 0.0 | 0.5% | 99.7% |
| hazards ×0.5 | 1.000 | 0.0 | 0.0 | 0.1% | 99.8% |
| hazards ×1.5 | 0.999 | 0.0 | 0.0 | 0.0% | 100.0% |
| without hazards | 0.999 | 0.0 | 0.0 | 0.3% | 99.7% |
Weight-sensitive factors (a ±50% change moves more than 10% of projects across a band): certification, country. They are present for almost every project and often among only two or three measured factors, so a low-coverage score is a coarse screen.
| Minimum coverage to publish a score | Projects with a headline score |
|---|---|
| 30% | 6,684 |
| 35% | 6,666 |
| 40% (live) | 6,283 |
| 45% | 2,959 |
| 50% | 2,925 |
| 60% | 2,721 |
Many published scores sit just above the 40% gate — filter on confidence, not only on the score.
Generated 2026-10-10 under methodology 1.2 over 12,954 rated projects.
Backtest
| Blind auc | 0.602 |
| Verdict | weak |
| Published auc | 0.829 |
| Blind auc with satellite | 0.477 |
Backtest run 2026-10-10T14:23:24Z.
Data sources and licences
What each layer is built from, on what terms, and how we use it — one table for every source, with the attribution each licence asks for: Data sources and licences. Registry data stays with the registries: we show identifying facts and link to the registry page; scores, grades and indicative prices are our own.
Methodology version 1.2. Scorecards and API responses carry the version they were produced under.
Use it
Every rating page has a printable scorecard (print it or save it as PDF from your browser) and the same data as JSON. Need the scores in bulk? Ask us.
The Veridio Score is Veridio's independent opinion based on public data. It is not investment advice.