The record
One scheme, every figure with its denominator, every dated action resolving to a stored span, and the absences stated in the body. An absent number means we could not support one, not that the number is zero.
credits carried against struck projects · denominator: the issuance workbook
39 named events · three carry a bare label
the market's own insurance reserve, sized for a hundred-year obligation · as at 9 December 2025
| Project | Class | Row | Quantity | Occurred | Became public | Evidence |
|---|
The data
Every count on this page resolves to these rows; verify a download against its sha256. The events file names its row kinds: exposure rows carry credit quantities; incident rows carry the incident identifier and its public date, with quantities on the struck project's exposure rows; disagreement rows carry magnitudes in their own columns and never sum with credits.
Built 1 September 2026. Sources: the CARB offset credit issuance workbook (updated 26 August 2026, sha256 59be0de20c15155f) and the forest buffer account balance series (November 2025 release, sha256 d4edec4b9f14cf67), both retrieved 28 August 2026 and mirrored. Every figure computes from those files; every dated action below resolves to a stored, verified span into the source's own text; the build replays from the repository.
01
Edition zero is the California compliance table: one scheme, chosen for the depth of its published record, project-level losses in structured form with an independent balance series to reconcile against. Edition one is compliance-anchored, adding the Australian scheme's verified relinquishments and a voluntary candidates register. Edition two is cross-scheme. Each rung is claimed when its coverage declaration can carry it, and not before.
02
39 dated incidents, carrying 335 span-verified dated actions, 12 September 2017 to 1 March 2026. An incident is the event; the project rows it strikes are exposures, listed separately below and never added to this count.
| Incident kind | Incidents |
|---|---|
| invalidation | 5 |
| reversal | 3 |
| reversal_intentional | 13 |
| reversal_termination | 4 |
| reversal_unintentional | 13 |
| termination | 1 |
| TOTAL | 39 |
Incidents and exposures link by project: an incident's identifier carries the struck project's id, so CAFR5225-A maps to CAFR5225's exposure rows, and the per-action mapping is in the events file. The counts stay separate because one incident strikes many rows and one project can suffer several incidents.
Dating mechanisms are the source's own: buffer-pool retirements to the day, operator and covered-entity surrenders and regulator invalidation actions to the month. Three incidents carry the bare label reversal because the source's prose does not say which kind; that is the file's truth and it stands as written.
03
| Class (issuer's label) | Projects | Credits |
|---|---|---|
| reversal_unintentional | 9 | 8,539,304 |
| reversal_intentional | 20 | 3,634,831 |
| invalidated_or_canceled | 5 | 134,083 |
Intentional reversal entries are negative quantities in the protocol's quantification; CARB confirmed by email on 1 September 2026 that some entries carry the minus notation and some let it be understood, an input inconsistency its next workbook update, expected 9 September 2026, should remove. The absolute figure is therefore the reversal quantity, and this row will be re-verified against the corrected workbook when it appears.
The 3 disagreement entries are ours, not the issuer's, and therefore do not sit in a table of the issuer's labels; they are recorded in their own section below.
04
Every unintentional reversal in the file is a forest project, so the ratio is published by project type: unintentional reversals are 3.71 per cent of the 230,106,914 credits issued to Forest-type projects (denominator: forest issuance, summed from the file at build time). Against all 293,101,089 credits of every project type the figure is 2.91 per cent, stated for scale only, because a forest numerator over an all-types denominator is a smaller number about nothing in particular. Buffer consumption is 8,967,106 retired of 36,228,767 contributed, 24.8 per cent as at 9 December 2025 (denominator: cumulative contributions). The buffer compensation timeline reconciles to the credit against the independent balance file on nine of ten attributable dates; one differs by 90 credits and three dates carry no attribution in the source's prose, stated below rather than smoothed.
05
The regulator's files disagree with themselves and with each other, and this register records that as it records any loss: the balance column implies 27,261,661 against a stated 27,254,352, three arithmetic breaks leaving the file 7,309 units out of agreement with itself; the issuance workbook and the buffer series disagree by 687,251 tonnes on contributions and 23,204 on retirements. Nobody holds both files and subtracts; this page does.
06
Occurred dates are stated only where the source states them; most exposures carry not stated, which is the file's truth. Three buffer retirement dates (36,083 credits in May 2015; 23,114 in December 2024; 280,582 of the October 2025 event) carry no incident attribution in the source's prose and are recorded as unattributed rather than guessed. The bare-label reversal incidents stay bare. An absent number here means we could not support one, not that the number is zero.
07
This edition covers one scheme, the California compliance offset programme, from its own published files, current to their stated dates.
Edition one will be compliance-anchored, and the voluntary layer will be a candidates register, never loss events. The two absences that force that design are findings, and they belong here in the body. First: the lawful bulk path to the voluntary registries carries 89 dated cancellation transactions across ten years (13.3 million credits at ACR, 1.2 million at the Climate Action Reserve, 71,100 at ART), and not one row states its cause; some of the largest are plausibly administrative conversions rather than losses. They will be published as candidates with cause not stated, each awaiting its deciding document, because publishing them as losses would manufacture the very numbers this register exists to check. Second: that same bulk path carries zero cancellation transactions for Verra and zero for Gold Standard, the two largest voluntary registries. Their loss events live in monitoring-report prose behind per-document retrieval, and they enter this register when that work is done and not before. A register that stated neither absence would be claiming coverage it does not have; these two sentences are the difference between this page and a marketing table.
The CDM transition is held at summary granularity and its figures are withheld pending verification at UNFCCC source. The Australian scheme's 21 relinquishment events (258,848 units, summed 1 September 2026 from the mirrored register) are verified and held for edition one's cross-scheme declaration. Material derived from the Climate Action Reserve's own store is internal-lane and appears nowhere in this artefact.
08
Corrections are new entries; the original stays visible. Any party named here has a right of reply, published alongside the entry: write to corrections@kyroq.com. Receipt is acknowledged within two UK business days and answered within ten UK business days.
The register is live and free to read. Leave your details and we will write when the next edition publishes.
Thank you. You are on the list, and we will write when a new edition publishes.
Independent reads are open. Leave your details and one line about the book, and we will reply.