Not a Failure of Will, but of Instrument Precursor Geometry in the 2023 Regional Bank Failures A Comparative Case Study of Silicon Valley Bank, Signature Bank, and First Republic Bank

This paper documents a single, bounded empirical question: did Silicon Valley Bank, Signature Bank, and First Republic Bank each exhibit a documented, multi-quarter deterioration signal prior to their 2023 failures? Using only primary sources — SEC filings (Form 10-K, 10-Q) for SVB, and FDIC Call Report data together with FDIC Office of Inspector General Material Loss Reviews for Signature Bank and First Republic Bank — the paper reconstructs each institution’s quarterly financial trajectory through 2022 and finds that, despite three structurally distinct proximate failure mechanisms (interest-rate and duration risk on an investment portfolio; crypto-sector deposit concentration compounded by contagion; and uninsured-deposit concentration compounded by an asset/liability duration mismatch), all three banks show a common structural shape: an internally observable, multi-quarter deterioration signal — declining interest-rate-shock absorption, eroding liquidity coverage, or breached internal risk-appetite thresholds — that was documented well before the eventual collapse and was not met with decisive corrective action. The paper is explicit that this is a descriptive finding, not a validated predictive one. It does not claim that these signals discriminate failing institutions from non-failing ones — establishing that would require a comparison panel of similarly situated banks that did not fail, evaluated against the same signals under a pre-registered detection threshold, which has not yet been undertaken. This limitation, along with data-resolution asymmetries across the three cases and the absence of a unified cross-institution measurement framework, is stated directly in Section 5 rather than left implicit. We believe that a defined precursor window in three separate reconstructions represents a significant advancement in coherence science. This paper is part of the Standard Coherence Fidelity Layer (SCFL) research program of the Upstream Coherence Measurement Stratum (UCMS), which develops substrate-agnostic measurement methods for detecting structural drift in complex coupled systems prior to visible failure. The banks’ own risk committees weren’t blind to their numbers — SVB tracked NII/EVE, First Republic tracked EVE against a stated risk-appetite policy, examiners tracked liquidity ratios. What none of them had was a framework that would tell them what those numbers, tracked together and compared to a rupture threshold, actually meant as a class of phenomenon. There was no concept of “precursor geometry” for them to check their data against — no prior cases assembled, no cross-institution comparison, no notion that a specific shape of multi-quarter deterioration is the kind of thing that recurs across different institutions and different failure mechanisms. The measurement apparatus that could recognize the pattern didn’t exist yet, because the pattern itself hadn’t been named or looked for across cases before.