Methodology

Every figure traces to an institution that stands behind it. Nothing is modelled where it can be observed, and nothing is asserted where the record is silent. The inputs fall into five groups.

• The carbon crediting registries. Project records, issuances, retirements and status changes come from the registries that hold them, including Verra, Gold Standard, the American Carbon Registry, the Climate Action Reserve and the Architecture for REDD+ Transactions. A university research group also maintains a consolidated record across registries, and it is used where a cross registry view is needed.
• Governments and their agencies. Emissions trading programmes publish their own allowance prices. The European Commission publishes its trading system and border adjustment rules. The United Kingdom government publishes the greenhouse gas conversion factors used for emissions reporting, and the United States environmental agency publishes spend based factors for supply chain estimates.
• Scientific and space agencies. NASA publishes satellite fire detections, which is how a forestry project's exposure to fire is checked against observation rather than assumption. The United States geological survey publishes earthquake records. Global forest monitoring programmes publish tree cover loss.
• International institutions. The United Nations publishes sustainable development indicators and the records of the climate convention. The World Bank publishes governance indicators, country economic series and natural hazard assessments. The science based targets initiative publishes sector decarbonisation pathways.
• Published statistics and the press. National grid carbon intensity comes from published energy statistics. Market coverage comes from carbon market publications and newswires.

Every dataset is collected on a schedule and read from our own copy, so a tool answers from a record that was already checked rather than from whatever a distant server happens to return while you wait. Each figure carries the date of the record it came from, because these sources move on very different clocks and presenting a quarterly figure as today's price would be its own kind of dishonesty.

Some of the most commonly quoted carbon numbers cannot be sourced without paying for them, and a free product that quotes them anyway is guessing.

• Voluntary transaction prices. What a voluntary credit actually traded at sits behind commercial price reporting services. No tool here states one. Where a voluntary credit's value is the question, the tools show what determines it instead: the methodology, the permanence class, the unretired overhang and the quality position.
• Per project impact claims. Jobs created, households benefited, hectares protected and litres of water improved are not published per project in any free dataset. They are named as unavailable rather than estimated.
• Anything on chain. Tokenised carbon pools and decentralised liquidity are not read, not priced and not counted.
• Broker and over the counter reports. Not obtainable, so not used, and no composite is built that pretends to include them.

The quality index is a weighted score across twelve dimensions. It is arithmetic on records, run the same way every time, so the same project scores the same today as it did last month unless the underlying record changed.

Methodology integrity carries the most weight because it is the question that decides whether a tonne was ever real. An independent integrity council assesses crediting methodologies against a set of core principles, and what it concluded about the methodology behind a project is used directly rather than interpreted.

The weighting is not fixed. A compliance buyer, who must surrender a credit to a regulator, can shift the weight onto compliance acceptance and verification currency. A removals buyer can shift it onto permanence and reversals. The weighting in force is shown alongside every score, and changing it changes the score visibly rather than silently.

A dimension with no evidence behind it is excluded from the score and reported as unscored, rather than being given a neutral middle value. Averaging in a guess would move the result while looking like information, and a project with thin public evidence should read as thin rather than as average.

There is no composite price and no blended index. Each compliance market's price is the price that market published, converted to a common currency where a comparison needs one, and shown with the date it was set.

Before any figure is computed from a price series, single day bad prints are removed. Some published series carry an isolated value at roughly half the surrounding level for exactly one day before returning, and a price that halves and doubles back overnight is a recording error rather than a market move. The number of prints removed is stated next to the market it was removed from.

Where a forward range is shown, it is each market's own realised trend carried forward, with a band set by that market's own realised volatility. It is a projection of what has already happened, described as an assumption rather than a forecast, and a market whose published prices are too sparse or too erratic to support one is listed with the reason instead of being given a line anyway.

A source going down is normal, and it must never turn into a wrong answer. Two rules cover it.

• Nothing fails silently. When a figure cannot be produced, the tool names which figure and why, and every other figure on the page continues to work. A blank space where a number should be is treated as a defect.
• Reports check themselves. A report that states a total also recomputes it from its own parts before it is shown. If the two disagree, nothing is shown at all. A figure that fails its own check is worse than no figure, because it looks like an answer.

Data is refreshed on a schedule rather than on demand, so a busy source cannot make a page slow and a failed refresh leaves the previous good copy in place with its own date attached.