Removals
The delivery ledger: what was contracted, what arrived.
Roughly 43 million tonnes of engineered removals contracted, about 1.2 per cent verifiably delivered. The number is real, widely misused, and the way it is usually calculated is wrong. What an honest delivery record requires.
The removals market has raised prices, signed offtakes measured in millions of tonnes and attracted the most sophisticated corporate buyers in the world. What it has not done is publish a delivery record. Work by researchers at the London School of Economics puts contracted volumes at roughly 43 million tonnes against about 1.2 per cent verifiably delivered by the end of 2025. That figure deserves care, and most of the care it deserves is in the direction of making it less alarming, not more. But it is the closest thing this market has to a delivery statistic, and the fact that it comes from a working paper rather than from any market institution is the point of this piece.
The number is not what most readers will assume
A one per cent delivery rate sounds like a market that has failed. It is not, and anyone who repeats it without the following paragraph is misusing it.
Most removal contracts are forward purchases with delivery scheduled years ahead. A contract signed in 2024 for delivery in 2029 has not failed to deliver; it is not due. Counting it in the denominator of a delivery ratio without saying so produces a number that is technically correct and analytically useless.
The statistical term for this is right censoring, and it is not an exotic technique. It is the standard machinery of survival analysis, used routinely in medicine, credit and insurance reserving. Treating a not-yet-due contract as a failure is the same error as counting a patient still alive at the end of a trial as a death.
Handled properly, the question is not what share of all contracted tonnes has arrived. It is what share of tonnes that were due has arrived, with everything not yet due held aside as censored and carried forward. That is a much smaller denominator, a much less dramatic headline, and the only version an actuary would accept.
We do not know the properly censored delivery ratio for this market. Neither does anyone else. That is the finding.
What is actually knowable today
It is worth being precise about which parts of this are hard and which are merely undone.
Issuance is public. The removal registries publish issued tonnes, machine-readable, with dates. That is a clean numerator for anything already delivered.
Contracts are semi-public. Several of the largest buyers disclose their purchases, some in considerable detail. Buyer disclosure is the single richest source in this market and it exists because a handful of purchasers decided transparency was worth more to them than discretion.
Schedules are mostly private. Here is where it breaks. Disclosure typically names the supplier, the pathway and the volume. It far less often names the delivery schedule. Without the schedule you cannot separate late from not-yet-due, which is precisely the separation the whole calculation turns on.
Supplier status is public but scattered. Company filings, insolvency notices and official gazettes record when a counterparty ceases trading. This data is excellent and nobody joins it to the contract record.
The deletions are the evidence
One finding from assembling this material is worth stating on its own, because it changes how the record has to be built.
When a removal supplier fails, its entries tend to disappear. Buyer portfolio pages are updated, the failed supplier is removed, and the public record quietly reflects a portfolio that no longer contains the failure. Nobody is concealing anything; a portfolio page is a description of current holdings, and a supplier that no longer exists is no longer a current holding.
The consequence is severe for anyone trying to measure this market. A researcher who scrapes buyer disclosures today sees a survivorship-filtered picture. The failures are not marked as failures. They are simply absent.
Which means the method has to change. You cannot photograph the record; you have to watch it. A weekly snapshot of every buyer disclosure, diffed against the previous week, converts an absence into an event: a supplier that was present and is now missing generates a candidate failure with a date attached, which can then be resolved against company filings and insolvency records.
Method note
- Snapshot buyer disclosures on a fixed weekly cadence and store every version.
- Diff each snapshot against the last. A removed supplier or contract emits a candidate event, dated to the week it disappeared.
- Resolve each candidate against company filings, insolvency registers and official notices before it is recorded as a supplier failure.
- Where the reason cannot be established from a public document, record the disappearance and abstain on the cause. An abstention with a date is worth more than a guess with a label.
Concentration, which is the other half of the risk
Delivery risk in this market is not evenly spread. Delivered tonnes are heavily concentrated in the pathways that were technologically ready first, which means the market's delivery track record is largely the track record of one or two technologies.
This matters for anyone underwriting delivery risk. A loss curve built on the delivered record is a loss curve for biochar with some other things attached. The pathways attracting the largest forward commitments, the geological and engineered routes, have the least delivery history and the highest capital intensity, which is precisely the combination that produces counterparty failure rather than late delivery.
What a delivery record has to contain
If somebody were to build this properly, and we are, the specification is not mysterious. Six fields, and the discipline to leave them empty when the evidence is absent.
| Field | Why it is load-bearing |
|---|---|
| Contracted volume | The denominator. Useless without the next row. |
| Delivery schedule | Separates late from not yet due. The field most often missing, and the one that decides whether the ratio means anything. |
| Delivered volume | Registry issuance, resolvable to a serial range. |
| Pathway | Because delivery risk is not comparable across technologies. |
| Supplier status | Trading, distressed, ceased. Joined from filings, not asserted. |
| Censoring flag | Explicit, per contract. A record that cannot say what it is holding aside cannot be trusted with what it counts. |
Two of those six are routinely public. Two are partly public. Two require joining sources nobody currently joins. None of it requires proprietary data, novel science or anyone's cooperation.
Why this is the most consequential gap in the market
Three reasons, in ascending order of importance.
The money has moved. Removals command prices an order of magnitude above avoidance credits, in some pathways two orders. Capital is being committed at those prices against a delivery record nobody has assembled.
The insurance is being written. Delivery and non-delivery cover is an active product line with real capacity behind it, and it is being priced from engineering first principles because there is no delivery table to price from. That is the position property underwriters were in before catastrophe loss data existed, and it resolved there only after the data was collected.
The failure mode is the severe one. A forest reversal is a partial loss with a physical cause you can observe. A removal that fails to deliver is a contract that did not happen, frequently against a counterparty that no longer exists, and frequently after payment. Severity concentrates exactly where the evidence is thinnest.
What we will publish, and what we will not
The Carbon Loss Register's first edition carries what the public record can currently support: delivery shortfalls where a schedule is disclosed, supplier failures resolved to filings, and the issuance record against contracted volumes where both are available. Every ratio will travel with its denominator and its censoring rule stated on the face of it.
What it will not carry is a headline delivery percentage for the market as a whole. Not because the number is unflattering, but because the schedules required to compute it honestly are not public, and publishing it anyway would make us exactly the kind of source this piece is arguing against.
Where the evidence does not support a view, the record will say so. That is a less satisfying artefact than a single arresting number. It is the only kind an underwriter can price from.