There is a building in London, just north of the Thames, where for more than seven centuries a specific kind of work has been done that almost no one thinks about. In the Goldsmiths' Hall, craftspeople sit at benches with instruments that weigh and assay metal. When a refiner or a jeweler brings an object in to be marked, the assayer scrapes a trace of material from it, tests the purity, records the result, and stamps the object with a small symbol. The symbol certifies that an independent party, paid only for the accuracy of the test, has verified what the object contains.

The London assay office was established in 1300. It has been continuously operating for seven hundred and twenty-six years. During that time, monarchies have risen and fallen, empires have been built and dismantled, currencies have collapsed, financial systems have been reinvented, and an entire industrial civilization has replaced the agrarian one that preceded it. The assay office is still there. It still stamps small symbols onto small objects. The reason it is still there - the reason every functioning gold market in the world has something like it - is that seven hundred and twenty-six years of experience have taught humanity that markets in valuable things do not work unless an independent party, with no stake in the outcome, performs the measurement.

The entire edifice of functioning markets rests on the ability of buyers and sellers to trust that what is being traded is actually what it is represented to be, and that trust is not produced by good intentions or professional ethics or regulatory oversight. It is produced by a physical, reproducible, independent measurement that anyone with equivalent equipment could replicate. Remove that foundation and the market does not gradually degrade. It collapses, usually in ways that take decades to fully understand, and always in the same direction: the party that controls the measurement extracts value from the parties that must take the measurement on faith.

I want to tell you something I believe to be true and that almost nobody in the field of ecological finance has been willing to say clearly. Every market that humanity has ever built to price living systems - to pay for the protection of forests, the preservation of biodiversity, the maintenance of watersheds, the provision of pollination services, the absorption of floodwater, the storage of atmospheric carbon - has been built without an assay office. Every one. For thirty years, we have been issuing certificates against representations we could not independently verify, certified by parties with direct interests in the outcome, traded through registries whose revenues scale with the volume of issuance, and retired in accounting exercises that no instrument ever watched happen. And for thirty years, we have been surprised, over and over, that the certificates turn out to be worth less than we thought.

They were not worth less than we thought. They were worth what the structure was capable of producing, which is nothing we could verify. The surprise was not in the result. The surprise was in how long it took to notice what the structure guaranteed.

This essay is about that structure, why it was adopted, why it failed, and what must replace it before the next generation of ecological markets - the ones being built right now, by well-intentioned people, on the same architecture that just collapsed - makes the same mistakes at a scale that will make the carbon market's failures look small. It is also about what I have been building for the last several years with a team that includes some of the most careful people I have ever worked with, and why I believe the next twenty-four months will determine whether the architecture of ecological finance shifts before mandatory disclosure regimes harden standards around the old one.

I am not writing as a disinterested analyst. I am writing as someone who is building the alternative, who believes the alternative is possible, and who thinks the field will spend another three decades failing if the architecture does not change. I am trying to persuade you. I am trying to persuade you with arguments and evidence, and I have tried to be honest about what remains unfinished and what could still go wrong. This is the most important thing I have ever written. I would like it to be worth reading.

The claim

The voluntary carbon market did not fail because its methodologies were weak, because its verifiers were captured, or because its participants were dishonest. It failed because of a structural property that no methodology, verifier, or participant could have fixed. The market was never a market for carbon. It was a market for representations of carbon, operating alongside an atmosphere that was completely indifferent to the representations.

The atmosphere does not read certificates. It responds to physical reality. A tonne of carbon dioxide that a project developer's consultant estimated to have been sequestered, that a seller-retained verifier reviewed and approved, that a registry issued as a credit, that a buyer retired as an offset against their emissions - none of that is a tonne of carbon. It is a paper claim about a tonne of carbon. The paper and the atmosphere operate in separate systems. The paper can be issued, traded, retired, accounted for, written into sustainability reports, and counted against net-zero commitments. The atmosphere continues to absorb whatever it is physically given, regardless of what the paper says.

This is what I mean by an unauditable market. Not that the audits were insufficient. Not that the auditors were compromised. But that the thing the market claimed to be trading had no independent measurement separating the claim from the underlying physical reality. The entire edifice rested on a chain of representations - project developer to verifier to registry to buyer - with no instrument, owned by any party outside that chain, ever measuring the physical thing.

Every study conducted on the voluntary carbon market over the last decade has reached the same conclusion. In October 2025, Oxford and Penn researchers published in the Annual Review of Environment and Resources the most comprehensive review ever attempted: twenty-five years of evidence, more than sixty empirical studies, every major project type. Their finding was that carbon offset programs routinely overestimated their climate impact by a factor of five to ten or more. Systematic. Deep-seated. Intractable. Incremental reform would not fix it. Most of the market should be phased out. A 2023 Guardian investigation of rainforest offsets from the largest certifier in the voluntary market found approximately ninety percent of examined credits were effectively phantom. A 2024 meta-analysis in Nature Communications found eighty-seven percent of offsets purchased by the world's twenty largest corporate buyers carried high risk of non-additionality. The voluntary carbon market contracted sixty percent between 2023 and 2025.

These results were not outliers. They were the expected output of an unauditable market. The surprise was not that the numbers were wrong. The surprise was how long it took to notice.

The distinction between verification and measurement

I want to name what I think the core psychological error has been, because I have made this error myself and I have seen the field make it for a quarter-century. The error is that we confuse verification with measurement.

Verification of an estimate is a judgment about whether the estimate was reasonable. Measurement is a physical fact. They are not the same kind of thing. The atmosphere responds to physical facts. It does not respond to judgments about reasonable estimates. The carbon market has been, for its entire operational history, a verification market pretending to be a measurement market, and the gap between verification and measurement is the gap in which every phantom credit was issued.

Consider what verification actually does. A project developer produces a model of how much carbon their project sequestered, using assumptions about the counterfactual baseline, extrapolation rules from sample plots to the full landscape, leakage estimates, permanence assumptions, and additionality judgments. A verifier, paid by the developer, reviews the model. The verifier examines the methodology, checks the inputs against the documentation, and issues an opinion on whether the developer's estimate was reasonable given the assumptions. If the verifier finds the estimate reasonable, the credits issue.

At no point in this process does any independent instrument, owned by any party outside the developer-verifier chain, measure any physical thing. The verifier is reviewing representations. Better verification produces more rigorous reviews of representations. It does not produce measurements. No amount of reform at the verification layer transforms an estimation market into a measurement market, because the verification layer and the measurement layer are different layers.

An auditable market is one in which the thing being traded can be independently measured by a party that does not benefit from the measurement outcome. The measurement is produced by instruments that are owned by the independent party. The measurement is continuous or reproducible at will. The raw data from which the measurement is derived is available for inspection, not just the summary statistics. The party that profits from higher numbers has no mechanism by which to produce higher numbers.

Gold markets are auditable. An assay office, paid for the assay and not for the result, receives a physical sample and runs a reproducible chemical analysis. The refinery cannot change the answer. The buyer can have the sample reassayed at any other assay office and get the same answer. The measurement is physical, independent, and reproducible. This is why gold markets have functioned, with variation only in ornament, for approximately two thousand years.

Equity markets are auditable at the measurement level that matters. The price at which a trade executed is recorded by an exchange that does not benefit from the price being higher or lower. The volume of shares outstanding is a public record the issuer cannot modify without disclosure. Quarterly earnings are audited by independent accounting firms whose professional licenses depend on the audit being accurate, subject to SEC enforcement and litigation risk. The arrangements are imperfect. When they fail - Enron, Worldcom, the 2008 structured credit rating failures - they fail visibly, attribution is possible, and the system reforms. That is what an auditable market looks like when it is working. The failures are detected, attributed, and corrected. The failure mode is reparative.

The carbon market's failure mode is not reparative. It is cumulative. Phantom credits issued in 2012 are not retrospectively unretired when a study in 2023 proves they did not represent real sequestration. The corporate buyer's net-zero claim remains on the record. The compensation paid to the project developer remains in their bank account. The atmospheric CO2 that was supposed to be offset was never offset. The market continues. The constituencies continue. The failure accumulates.

This is what happens when you replace measurement with verification and hope the replacement is adequate. The replacement is not adequate. It has never been adequate. It is not adequate now.

Will continue...

Alex Roessner, Co-founder, Landseed PBC. E-mail- alex.roessner@landseed.earth