The 2022 UK gilt crisis: public evidence, funding identification and conditional policy counterfactuals

We examine the 2022 UK gilt crisis using public Bank of England yield curves and emergency purchase records. The thirty-year nominal spot yield rose 112.84 basis points from 22 to 27 September and fell 112.83 basis points on 28 September. Purchase proceeds totalled 19.259 billion pounds. These observations measure changes around events; identifying a causal purchase response requires further restrictions. We derive the exact coefficient set under a stated bound on omitted daily disturbances and distinguish this sensitivity calculation from sampling inference. A funded position game and a collateral ledger state which additional records would identify individual constraints. We then connect observed yields and purchases to a conditional response model. Each assumed parameter vector reproduces the observed path through its own fitted residuals, yet removing execution flows gives different counterfactual paths. The finite scenario range measures sensitivity to those assumptions. Public prices and purchases alone leave the purchase coefficient and individual collateral exposure unrestricted when structural disturbances are unrestricted. The empirical contribution is a documented public-data reconstruction and an explicit account of the observations needed to distinguish mechanisms. This deposit contains an English preprint, a supplementary Japanese explanatory report, and research materials including manuscript sources, author-written code, processed observations, provenance records and synthetic validation examples. The Japanese report is explanatory material rather than a word-for-word translation. Public-data observations, imposed disturbance allowances and illustrative structural assumptions are distinguished throughout. The finite-grid execution-removal comparisons are conditional sensitivity calculations, not historical causal estimates or confidence intervals. The original Bank of England XLSX workbooks are not redistributed in this public deposit. Their official retrieval locations, recorded vintage and SHA-256 hashes are provided. Downstream calculations can be rerun from the included processed tables. Re-extraction from the source workbooks requires obtaining the four files separately and matching the recorded hashes; a later provider vintage is not a bitwise replication of the November 2026 inputs. Component-specific rights are explained in README_Truss_Deposit.txt and RIGHTS_AND_REUSE.txt in the research archive. The manuscript and source-data vintage are dated 2 November 2026. The public-deposit packaging and data-availability clarification were prepared on 3 November 2026. No external funding supported the work. The author declares no competing interests. AI-assisted tools supported drafting, code preparation and language editing; the author remains responsible for the final analysis and verification.