The K-A-T Framework: A Unified Framework for Macro-Financial Propagation Through Funding Constraints, Networks, and Policy Backstops

Version 2 (2026-06) — supersedes v1 (citable at its version DOI); corrections only. (1) The Section-5 meta-analysis heading is corrected to Random-Effects (DerSimonian-Laird) to match the reported method (the body and abstract already report a frequentist DerSimonian-Laird random-effects synthesis, not a hierarchical Bayesian model). (2) The synthesis pools ELEVEN in-sample T-shock episodes (1999-2025); grand-mean half-life compression -72.0% (95% CI [-82.6%, -55.1%], tau^2=0.44). (3) The panel LP-IV is honestly weak-identified — first stages are weak, so identification rests on Anderson-Rubin weak-IV-robust confidence sets and literature triangulation, not a first-stage-F headline. No new claims.

Theoretical companion to the K-A-T monolith (Stanisljevic 2026d, SSRN Abstract ID 6668858). The paper proposes K-A-T as a structural framework with three channels (funding constraints K, network amplification A, policy backstop T) and derives Proposition 1: HL_i = ln 2 / (lambda_T + mu_i), with lambda_T proportional to T. As T -> 0, HL remains finite, bounded above by the natural decay rate, rejecting unbounded contagion in the absence of policy intervention. Validation via hierarchical Bayesian random-effects meta-analysis of ten historical T-shock episodes spanning four decades (Discount Window 2008, ECB SMP 2010, Operation Twist 2011, OMT 2012, USD swap-line 2020, FIMA 2020, SRF 2021, BoE LDI 2022, TPI 2022, BTFP 2023, SRF cap removal 2025). Grand-mean compression -64.9% (95% CI [-75.5%, -49.8%], p<0.0001, tau^2 = 0.202). Sensitivity to heterogeneity estimator (DerSimonian-Laird vs REML vs Paule-Mandel) shifts grand-mean by less than 1pp. Leave-one-out analysis confirms no single episode drives the result. The Feb-Apr 2026 private credit gate cascade is a held-out posterior predictive check; realized -73.0% lies inside the 95% predictive interval [-86.4%, -9.1%].