<p>We propose and test an empirical framework for macro-financial propagation—K-A-T—that links (i) funding and collateral constraints (K), proxied by USD cross-currency basis deviations and high-frequency monetary surprises, (ii) network amplification (A), via eigenvector centrality on international input-output tables and, new in this paper, a stylized GP/BDC-bank financial adjacency matrix constructed from EDGAR 10-Q filings, and (iii) policy backstop credibility (T), via OMT/TPI in the euro area and SRF/FIMA in the United States. We formalize K-A-T in a two-period model with a put-option representation of the backstop. A peer-reviewed event-study triangulation (Altavilla-Giannone-Lenza 2016 for OMT; NY Fed 2025 for SRF cap removal; MEF 2026 for BTP-Bund) documents half-life compressions of 62% (OMT), 75% (cumulative post-TPI), and intraday containment (SRF).</p> <p>The February 19–April 8, 2026 semi-liquid private credit gate wave—eight funds totalling $178 billion in AUM gated at 45–100% fill rates, with JPMorgan marking down $22.2 billion in collateral and Deutsche Bank disclosing a €26 billion portfolio—provides the central natural experiment. Using daily returns for 21 firms over the 35-trading-day window, exposed alternative asset managers accumulated a CAR of −10.1% at day +35 versus +0.1% for low-exposure diversified financials (difference −10.2 pp, permutation p = 0.087). Business Development Company returns over the event window have an R² of only 18.8% from public-credit comparators (LSTA, HY, IG), identifying a 19.4%-annualized K-channel-specific volatility component. A triangulation across Proskauer, S&amp;P, Fitch, and KBRA yields a Q4 2025 private-credit default rate of 4.7% (interquartile range 2.5–5.8%), rather than the selection-biased 9.4% Fitch privately-monitored figure widely cited. A stylized financial-network stress test calibrated to real BDC balance sheets and the Berrospide et al. (2025) G-SIB bank-lending shares shows that moderate contagion scenarios (up to 25% default with 50% LGD on $1.34 trillion bank-to-PC lending) remain below the Fed's 2025 DFAST CET1 drawdown of 1.6 pp; only catastrophic (40% / 60%) scenarios breach the baseline. Consistent with our model's Proposition 1(iv), the April 2026 data show that stress under T ≈ 0 has a longer but finite half-life, not unbounded escalation.</p> <p>We propose four narrowly-targeted backstop-architecture reforms calibrated to these estimates and designed to minimize moral hazard.</p>