This record contains Document 03 of the DFG Empirical Programme submission package: Paper B-H4 — Substrate-Matched Financial Critical Slowing Down: A Cross-Crisis Test of DFG H4-β. This synthesis paper tests the DFG H4-β prediction that financial crises are preceded by critical-slowing-down-style early-warning signatures, operationalised primarily as rising rolling-window variance in stress observables. The paper refines the original H4-β claim into a bounded, substrate-matched form: financial CSD is most reliably detectable when the stress observable’s channel matches the crisis substrate. The paper evaluates seven major crises from 1987 to 2020 across 13 source series / 18 observable transforms spanning US credit, US bond and volatility indicators, EU systemic stress, China FX, and China-linked macro-financial indicators. It includes null-window false-positive controls and distinguishes matched-observable detection from unmatched-observable failure. The central finding is that 7/7 crises are detected at ≥75% under matched observables and windows, while the apparent failures of Black Monday 1987, EZ Debt 2011, and China 2015 are reclassified as substrate-mismatch or window-mismatch artefacts. Six of seven crises show stronger specificity, while Black Monday is treated as compatible with lower specificity confidence due to its shorter one-year buildup window and higher null false-positive rate. The bounded conclusion is that DFG H4-β is not a universal one-indicator financial warning law. It is a regime-bounded prediction: financial CSD is detectable when stress observable and crisis substrate match, and may fail when they mismatch. The detected variance trends are interpreted as CSD-compatible / fold-like buildup signatures, not as proof of strict fold-bifurcation dynamics. This record is part of the DFG Empirical Programme submission package. It assumes the regime-atlas methodology established in Document 01, the Compatibility Atlas v1.3. Cell-level incompatibility should be interpreted as a regime-boundary marker rather than theory-level falsification.
Bin Seol (Sat,) studied this question.