Scoring Methodology
CarbonMeld assigns a score from 0 to 10 to each project based on public registry documents and marketplace metadata. The analysis is generated using CarbonMeld's internal scoring framework and AI-assisted review pipeline, following the criteria defined on this page. Starting from v2.0, each project is classified by sector and evaluated with sector-specific penalty gates in addition to the universal scoring framework.
Four Dimensions, Weights and Criteria
| Dimension | Weight | What is assessed |
|---|---|---|
| Integrity | 35% | Additionality, quality and conservatism of baseline, permanence risk, leakage |
| Transparency | 25% | Public availability of data, accessible MRV, public PDD, findable official registry record |
| Claim Safety | 25% | Greenwashing risk for the buyer, ICVCM/CCP eligibility, CORSIA eligibility |
| Documentation | 15% | Availability of PDD, monitoring reports, third-party verification reports |
overall = integrity × 0.35 + transparency × 0.25 + claim_safety × 0.25 + documentation × 0.15
Weights are uniform across all sectors. Sector-specific differentiation is achieved through penalty gates (see below), not through different weights.
Operational Scale
| Range | Operational meaning |
|---|---|
| 9 – 10 | Strong primary evidence, complete documentation, baseline independently verified |
| 7 – 8.5 | Documentation available, minor concerns or secondary uncertainties |
| 5 – 6.5 | Partial data, at least one area with weak or absent evidence |
| 3 – 4.5 | Insufficient evidence across multiple dimensions, possible structural issues |
| 0 – 2.5 | Serious red flags, missing documentation, material risks identified |
Sector Classification
Each project is automatically classified into one of the following sectors using a deterministic keyword-matching system. Classification is based on the project's credit type, name, and description — checked in that priority order. No AI model is involved in classification.
| Sector | Examples |
|---|---|
| redd | REDD+, avoided deforestation, forest protection |
| arr | Afforestation, reforestation, revegetation |
| cookstoves | Clean cooking, improved cookstoves, fuel-efficient stoves |
| biogas | Biogas digesters, biogas development |
| biomass | Biomass cogeneration, rice husk, bio-oil, wood pellet |
| renewable_energy | Wind, solar, hydro, geothermal, run-of-river |
| methane | Landfill gas, coal mine methane, methane destruction |
| biochar | Biochar production and application |
| dac | Direct air capture |
| blue_carbon | Mangroves, seagrass, saltmarsh, coastal wetlands |
| soil_carbon | Carbon farming, agroforestry, agricultural land management |
| water | Safe water access, water purification, boreholes |
| industrial | Energy efficiency, waste heat recovery, fuel switching, composting, waste-to-energy |
| other | Projects that do not match any of the above |
Each project page displays which sector classification was applied.
Combination Rules and Gates
The linear formula can be too lenient when one dimension has a serious problem. Gates are applied deterministically after the weighted average is computed.
Universal gates
If Integrity is below 4.0, the overall score is capped at 6.0. A structural additionality problem cannot be masked by strong documentation.
If Documentation is below 3.0, the Transparency score is reduced by 1.0. Transparency depends partly on access to underlying documents. When documentation is severely limited, the transparency score is constrained accordingly.
If the additionality test is present but not confirmed by a Validation/Verification Body, and the additionality semaphore is green, it is downgraded to yellow. This does not affect the numerical score.
Sector-specific penalty gates
These gates apply additional penalties to projects in specific sectors where certain risks are structurally important. Each gate reads structured data extracted from project documents. Gates that check for absent information only fire when project documents are available — a project with no uploaded documents is not penalized for missing fields.
If leakage is not addressed or not documented in available project documents, Integrity is reduced by 1.5 points. For REDD+ and ARR projects, the absence of any leakage documentation is treated as a structural red flag — unlike sectors such as biochar where leakage is not a primary concern.
If the buffer pool percentage is not disclosed in available documents, a red flag is added. No score penalty is applied — the AI evaluator already factors buffer pool adequacy into its assessment.
If the fraction of non-renewable biomass (fNRB) is based on a national default value rather than a project-specific assessment, Claim Safety is reduced by 1.0 point. National defaults are known to overestimate fNRB in many regions.
If usage monitoring is assumed or not stated (rather than measured through smart meters, surveys, or self-reporting), Claim Safety is reduced by 0.5 points. These two cookstove/biogas gates can stack: national-default fNRB and no usage monitoring together produce a total penalty of −1.5 on Claim Safety.
If no additionality test type is documented, Integrity is reduced by 1.0 point. For renewable energy projects, the absence of an additionality demonstration is the primary risk — unlike other sectors where a missing field may simply indicate incomplete document extraction.
Sectors not listed above (biomass, methane, biochar, dac, blue_carbon, soil_carbon, water, industrial, other) are classified for informational purposes and receive the universal gates only. No sector-specific penalty gates apply to these sectors in v2.0.
Gate application order
1. AI-generated sub-scores are validated and adjusted: the documentation gate is applied, overall is recomputed from adjusted sub-scores, and the integrity cap is enforced.
2. Sector-specific penalty gates are applied sequentially. After each gate that modifies a sub-score, overall is recomputed and the integrity cap is re-enforced.
3. All sub-scores are clamped to [0.0, 10.0] before each recomputation.
Each project page shows which gates were applied under the "Gates Applied" field.
Vintage Assessment
Each project receives a vintage flag based on the recency of its monitoring data. The latest year mentioned in the monitoring period is compared to the current year.
| Flag | Condition | Effect |
|---|---|---|
| recent | Monitoring data less than 3 years old | No effect |
| aging | Monitoring data 3–5 years old | No effect on score |
| stale | Monitoring data more than 5 years old | Confidence downgraded by one level |
If monitoring period data is not available, the project defaults to "recent" — no penalty is applied for unknown vintage. The vintage flag affects confidence only, never the numerical score. A project with stale data may still score well on all dimensions, but the confidence level signals that the underlying evidence is dated.
Semaphore Indicators
Each project is assessed on six risk indicators, each shown as a traffic-light signal. Semaphore indicators are qualitative assessments shown alongside the numerical score. They help identify specific risk areas that the overall score alone may not highlight.
| Indicator | What it measures |
|---|---|
| Additionality | Is there evidence the project would not have happened without carbon finance? |
| Permanence | How robust is the mechanism preventing reversal of emission reductions? |
| Leakage | Is emissions displacement outside the project boundary accounted for? |
| Baseline | Is the emissions baseline conservative and well-documented? |
| Safeguards | Are environmental and social safeguards in place and monitored? |
| Double claim | Is there risk of the same reduction being claimed by multiple parties? |
Evidence Scope and Confidence
Each score is assessed against the evidence actually available. Evidence scope defines what categories of documents and data were accessible at the time of scoring. Confidence is derived from scope, gate coverage, and vintage, and reflects how much weight the resulting score can bear.
Evidence scope categories
| Scope | What is included |
|---|---|
| Registry-grade | Official registry record, PDD, third-party verification report, and at least one monitoring report, all publicly accessible |
| Partial | Registry record present but one or more key documents absent, outdated, or inaccessible at the time of scoring |
| Marketplace-only | Marketplace listing metadata only, no linked registry record or project documents available |
Confidence levels
| Confidence | Meaning |
|---|---|
| high | Verified registry documents present, complete data, score reflects primary evidence |
| medium | Some documents absent or not recent, score reflects available evidence with noted gaps |
| low | No registry documents available, or vintage flag is stale with medium base confidence |
Confidence is computed deterministically from document availability, gate verifiability, and vintage status. It is not produced by the AI model. A 6.8 with high confidence is not equivalent to a 6.8 with low confidence. Confidence does not adjust the score numerically — it qualifies the reliability of the evidence behind it.
What Would Improve
Each project audit includes a "what would improve" field listing specific actions that could raise the project's score. Examples include: uploading a more recent monitoring report, documenting fNRB with a project-specific assessment, or providing VVB confirmation of the additionality test. This field is generated by the AI evaluator based on gaps identified during the audit.
Official Sources Used
- Verra Registry (VCS), project page, issuance reports, PDD
- Gold Standard Registry, project page and documentation
- CDM Registry (UNFCCC), for CDM projects
- CORSIA database (ICAO), for CORSIA eligibility
- Marketplace metadata (ClimateTrade, Carbonmark, Toucan, and others)
- Third-party verification documents downloaded from registries
What Is Not Considered
- Market price — the score does not account for the credit price
- Buyer or seller reputation
- Unofficial third-party opinions or reviews
- Non-public communications with project developers
- On-chain transaction history for tokenised credits
A score is not a certification, financial recommendation, or guarantee of environmental outcome.
Version Log
| Version | Date | Changes |
|---|---|---|
| v2.0 | 2026-03-29 | Sector classification (14 sectors). Sector-specific penalty gates for REDD+/ARR, cookstoves/biogas, and renewable energy. Vintage assessment with confidence downgrade. Semaphore indicators documented. "What would improve" field documented. |
| v1.0 | 2026-03-16 | First public version. 4 dimensions, weights 35/25/25/15, Integrity gate and Documentation gate. |
Governance
The scores published in the database are generated by a standardized pipeline applied to documentary inputs. Commercial relationships, advertising, and consulting activity do not modify the methodology criteria and do not guarantee specific results for any project.
Methodology updates are versioned and logged above. The current active version is applied uniformly to all projects in the database at the time of each pipeline run.
See also: Editorial Policy · About the System
Previous Methodology Versions
v1.0 — General Methodology (2026-03-16)
Overview
Version 1.0 used a common rubric for all project types. This was a deliberate choice to ensure comparability, but introduced limitations: sector-specific risks were underrepresented.
Dimensions and Weights
Identical to v2.0: Integrity (35%), Transparency (25%), Claim Safety (25%), Documentation (15%).
Gates
Only two universal gates were applied. No sector-specific gates existed.
If Integrity below 4.0, overall capped at 6.0.
If Documentation below 3.0, Transparency reduced by 1.0.
Hard and Discretionary Criteria
Hard criteria were objectively verifiable conditions with a defined weight: public PDD present limits Documentation ceiling; third-party verification report limits Documentation ceiling; findable registry record is positive for Transparency; CORSIA eligibility is positive for Claim Safety. Discretionary criteria — baseline quality, additionality strength, permanence risk, leakage quantification — were evaluative and did not express as fixed increments.
Project Types and Limitations
| Project type | Primary undifferentiated risk |
|---|---|
| REDD+ / Forest Carbon | Permanence is critical; weight did not fully capture sector specificity |
| ARR (Afforestation/Reforestation) | Additionality requires demonstration of counterfactual land use |
| Biochar | Century-scale permanence is difficult to verify with standard criteria |
| DAC (Direct Air Capture) | Greenwashing risk from "permanent removal" claims |
| Methane destruction | Additionality tied to local regulation |
| Cookstoves | Energy consumption baseline difficult to verify in the field |
What Changed in v2.0
- Projects are now classified into 14 sectors, enabling targeted risk assessment
- Sector-specific penalty gates reduce scores when key documentation is absent or inadequate
- Vintage assessment tracks recency of monitoring data and adjusts confidence accordingly
- Semaphore indicators and "what would improve" fields are now documented in the methodology
- Confidence is computed deterministically from document availability, gate coverage, and vintage — not by the AI model