Permanence
The buffer account, reconciled.
The market’s largest insurance reserve is 24.8 per cent consumed, thirteen years into a hundred-year promise. The figures have been public throughout. Nobody had added them up.
There is an account in the carbon market that nobody reconciles. It is the largest insurance reserve the industry has, it underwrites the permanence claim attached to every credit a compliance buyer holds, and its balance is published by a regulator in tables anyone can download. We added it up. Of 36.23 million tonnes ever contributed to California's forest carbon buffer pool, 8.97 million have been retired against reversals. That is 24.8 per cent of the reserve, consumed roughly thirteen years into a promise that runs for a hundred.
This is not a modelling result and it is not an estimate. It is arithmetic over a public table. The remarkable thing is not the number; it is that the number has been sitting in the open for years and has never, so far as any survey of the public record can establish, been assembled and stated with its denominator.
A note on the source, and why it matters
CARB's own file does not reconcile. Contributions sum to 36,228,767 and retirements to 8,967,106, which implies a closing balance of 27,261,661. The file states 27,254,352, a difference of 7,309 units. Walking the running balance row by row isolates the cause exactly: three arithmetic breaks in the regulator's own balance column, on 23 December 2024, 8 April 2025 and 22 April 2025, where one retirement was reversed out of the balance and two additions were never carried into it. Every row after 22 April 2025 inherits the corrupted base.
We publish both readings rather than choosing one, because a reader is entitled to see the discrepancy rather than a number we have quietly smoothed. The ratio is 24.8 per cent on either basis, so the conclusion is unaffected. This is what the register is for: the figures were public for years, and nobody had added them up far enough to notice.
What the buffer is actually promising
A forest carbon credit is a claim about the future. Under California's compliance programme, permanence has been interpreted to mean carbon kept out of the atmosphere for a minimum of one hundred years. No certificate can underwrite a century. Trees burn, get sick, and are cut down, and atmospheric carbon dioxide outlasts any individual stand by a wide margin.
The market squares that circle not by promising each project will survive, but by self-insuring the portfolio. Every project contributes a share of its issued credits to a communal reserve. When a project suffers an unintentional reversal, credits are retired from the pool to cover the loss, one buffer credit per tonne released. As long as the pool stays solvent, the permanence claim attached to every credit in circulation holds.
It is, in structure, a mutual. Every participant contributes to a shared fund against a peril none of them can individually withstand. Mutuals are a serious and well understood instrument, and there is nothing wrong with the design. But a mutual has an actuary, publishes a solvency position, and reconciles claims against contributions on a schedule. This one has none of those things.
The premium is a schedule, not a price
Look closely at how the contribution is calculated and one thing becomes obvious to anyone who has priced risk for a living. The rates are administrative categories, not underwriting.
Wildfire risk draws a contribution of two to four per cent of credited carbon, lower for projects claiming active fire management. Disease and insects take a fixed three per cent. Other catastrophic natural risks, wind, ice and flood, take another three per cent. Financial and management risks, including bankruptcy, take one to nine per cent. Total contributions run from 8.7 to 19.2 per cent.
Every one of those numbers is flat. A project in a basin with a five-year fire return interval contributes the same wildfire share as a project in a temperate zone that has not burned in a century, adjusted only by a management claim the project itself asserts. In any other line of insurance this would be described as a rating plan with no rating factors.
The buffer pool charges a fixed premium for a variable peril, and nobody publishes the loss ratio. That is not a criticism of the people who designed it. It is a description of a mutual that was built before anyone had the data to price it, and which has never acquired the data since.
The run rate, stated plainly
Here is the arithmetic that should concern anyone holding these credits. If the reserve is 24.8 per cent consumed after thirteen years, the observed depletion rate is roughly 1.91 per cent of the original pool per year. The design implicitly assumes something closer to one per cent, because that is the rate at which a fully contributed reserve survives a hundred-year obligation.
At the observed rate, the reserve is exhausted in the mid-fifties of the promise, not at the end of it.
Three caveats, and we state them because a figure without its caveats is advocacy rather than analysis.
The pool is still being contributed to. New projects add credits as they are issued, so the denominator grows. A depletion rate against a growing base is not the same as a depletion rate against a fixed one, and the honest reading is that the reserve is being drawn faster than it is being replenished in loss years, not that it faces a fixed countdown.
Linear extrapolation is the crudest possible method. It assumes the next forty years look like the last thirteen. There is a serious argument that they will look worse, because fire return intervals are shortening in exactly the geographies where these projects concentrate. There is also an argument they will look better, if contribution rates rise or project selection improves. We are not modelling either. We are stating the rate that has actually been observed.
Not every retirement is a catastrophe. The pool covers unintentional reversals, and the classification of what counts as unintentional is itself contested. Some retirements reflect clerical correction rather than carbon returning to the atmosphere. This is precisely why an event record needs a cause taxonomy rather than a single total, and why our register separates unintentional reversal from intentional reversal from administrative cancellation, with sub-causes beneath each.
Why nobody has said this before
The data has been public throughout. So the interesting question is not what the number is, but why a market with several well-funded rating agencies, a growing insurance sector and an active academic literature has not maintained a running reconciliation of its own largest reserve.
The answer, we think, is structural rather than negligent. Consider who might have done it.
| Party | Why they have not |
|---|---|
| The programme administrator | Publishes the tables, which is the job. Reconciling them into a solvency narrative about its own instrument is a different job with obvious institutional friction. |
| Rating agencies | Sell a view formed before the fact about whether a credit is likely to be sound. A reserve depletion statistic says nothing about any individual project and cannot be sold per project. |
| Project developers | Contribute to the pool and benefit from confidence in it. The incentive runs the wrong way. |
| Insurers | Would benefit, and several are now writing reversal and buffer cover. But their own claims experience is confidential by construction, and their analytics answer what is true now rather than what happened before. |
| Academics | Have done the hardest part of this work, repeatedly and well. What academic incentives do not reward is maintaining a series, indefinitely, after the paper is published. |
That last row is the one worth sitting with. The underlying research is excellent. What is missing is not intelligence or method. It is the unglamorous, permanent obligation to keep the account current, which is a job nobody in this market has ever been paid to hold.
What it means if you are holding these credits
Three consequences follow, and they land on three different desks.
If you underwrite reversal risk, the buffer is your competitor and your backstop simultaneously. Cover priced above a buffer that is depleting faster than its design assumed is cover priced against a moving attachment point. The pool's solvency is an input to your pricing and nobody is publishing it.
If you mark these positions, the permanence claim is doing work in your valuation whether or not you have made it explicit. A hundred-year claim backed by a reserve on a fifty-two-year run rate is a different asset from a hundred-year claim backed by a reserve tracking to plan. Your auditor is entitled to ask which one you hold.
If you are lending against them, the collateral's value depends on a mutual whose accounts nobody publishes. That is an unusual thing to accept in secured lending, and it is accepted here only because the alternative, which is an independent reconciliation, did not exist.
What we did, so you can repeat it
- Downloaded the regulator's issuance and reversal tables directly, rather than a derivative dataset.
- Summed contributions to the reserve across all vintages to date. Denominator: 36.23 Mt.
- Summed retirements from the reserve against reversals. Numerator: 8.97 Mt.
- Stated the ratio with both figures and the as-at date attached, because a percentage without its denominator cannot be checked.
- Declined to model forward. The extrapolation shown is linear and labelled as such.
The point is not the number
It would be easy to read this as an argument that the buffer pool is failing. That is not the claim, and the evidence does not support it. Reserves are drawn down; that is what they are for. A pool consumed at 24.8 per cent after thirteen years might be perfectly adequate if contributions continue to grow and the loss years just observed prove unrepresentative.
The claim is narrower and, we think, harder to argue with. Nobody knows whether it is adequate, because nobody is keeping the account. That is a solvable problem, it requires no new science, and the data has been public the entire time.
Every other market that pooled a risk eventually appointed somebody to reconcile the pool and publish the position. Carbon has not yet. We are building the record that makes it possible, starting here, and publishing it free, because a reserve that everyone relies on should have an account that everyone can read.