How the panel is built

Methods

The SPI panel is built from the full text of three federal bank supervisors' written records: the Office of the Comptroller of the Currency (annual reports 1863–present, comptroller's handbook, interpretive letters), the Federal Reserve (annual reports 1914–present, supervisory letters, examination manuals), and the Federal Deposit Insurance Corporation (annual reports 1934–present, supervisory insights, examiner manuals). Each document is scored against a domain taxonomy refined over multiple iterations and validated against forensic episodes whose timing is known.

What the methods chapter will contain

• The corpus, document by document, with provenance.
• The L0 / L1 taxonomy and how it was constructed.
• The scoring procedure and validation against ground-truth episodes.
• Inference choices: clustering, sample restrictions, robustness checks. In particular, why agency-head clustering (G ≈ 75) is the headline test for partisan inference and why a G=3 wild cluster bootstrap is not.
• Replication code and data versioning.