The California compliance record does not only say how much has reversed. It says the loss does not fall evenly, and the way it concentrates is the part a book has to price.
What this rests on, said plainly. These figures come from the regulator's own structured issuance and buffer-account data, the same files edition zero is built from, not from a separate determination document for each reversal. That is a real limit: the concentration is only as good as the regulator's bookkeeping, which this register has already shown does not fully reconcile with itself. It is stated here, in the open, because it is what a careful reader should test before anything else.
Across the dated reversal actions in edition zero, 19,192,941 credits are booked as reversed, resolved to 26 projects. The distribution is steep. The single largest project is 56.2 per cent of all reversed credits. The top three are 68.3 per cent. The top five are 78.6 per cent. Nine of the twenty-six projects hold ninety per cent of the total.
Reversal here is not a background rate. It is a small number of large, correlated events.
Timing concentrates too. A single year, 2023, carries 42 per cent of the reversed credits, the signature of one fire season rather than a steady rate. And the mechanism is buffer-dominated: of the dated reversal actions, 249 draw the shared buffer pool and 54 are operator surrender. The losses fall on the pooled reserve that is meant to make the permanence promise credible across the whole programme, before they reach a holder.
| Rank | Project | Credits reversed | Share | Cumulative |
|---|---|---|---|---|
| 1 | CAFR5225 | 10,785,154 | 56.2% | 56.2% |
| 2 | CAFR5038 | 1,157,566 | 6.0% | 62.2% |
| 3 | CAFR5058 | 1,157,566 | 6.0% | 68.3% |
| 4 | CAFR5233 | 1,145,363 | 6.0% | 74.2% |
| 5 | CAFR5076 | 847,895 | 4.4% | 78.6% |
The twenty-one further projects to one hundred per cent are in the data file below.
Why it matters
A flat buffer percentage, set once by risk category, does not price this shape. An insurance policy replacing that buffer has to, because a correlated tail is exactly what an attachment point and a limit are for. The record does not forecast which project fails next; it says the tail is heavy and correlated, which is a reason to price correlation rather than to sum independent probabilities.
One scheme travels with this, California compliance forest, the jurisdiction where the record is richest; the cross-scheme picture widens as further editions land. And concentration measured on a realised record describes what has happened, not what will. Every share above has the same denominator, 19,192,941 reversed credits.
Provenance, and how to check it
Computed from edition zero, the California table (artefact sha256 d7db0f55823b192ada3f95c1143a0e6e7826bedfd5f16685ffd23762f6cf4b91), incident series. This figure is derived here; it is not published inside the edition, so it sits outside the edition's hash chain. To replay it from the published file, download the incident series from the edition zero page and run:
python3 concentration_recompute.py incident-series.csv
When edition one supersedes edition zero, run the same command against the new file. A different answer is the record having changed, which is the point, not this article having been wrong.
Download the ranked data: twenty-six projects with cumulative share, from the edition-zero incident series, spans-verified, computed 2 September 2026.