A simple question drives this paper: why does the same debt-to-GDP ratio imply very different sovereign risk across countries and over time? I build a country-year panel of 195 economies for 1980-2024 that merges the Bank of Canada-Bank of England Sovereign Default Database with IMF and World Bank macro-fiscal series, and I evaluate early-warning models strictly out of sample under a horizon-purged time split: covariates are lagged one year, models are fitted through 2009, isotonic-recalibrated on a window that ends before the test labels begin, and tested on the 2017-2024 default wave. Defaults are measured as materiality crossings with hysteresis rather than any positive stock. The debt ratio alone is a weak predictor: AUROC 0.71 one year ahead, pseudo-R2 0.02 in sample. Adding fiscal, macro, liquidity and currency-structure variables lifts out-of-sample AUROC to 0.84 one year ahead and 0.79-0.80 five years ahead; the paired bootstrap improvement over the fiscal-macro specification is positive in 96-98 percent of 1,000 full-pipeline resamples, and Brier scores beat both ex-ante and oracle climatologies at every horizon. The headline conditional story does not survive: neither the debt x (r-g) interaction nor an out-of-sample subgroup split detects the snowball channel in this design. I then embed the estimated stress model - re-estimated excluding France - in a stochastic debt-dynamics engine built on the security-level ledger of Agence France Tresor and simulate 2,000 paths per scenario to 2050. France's 2026 configuration maps to a stress index of 0.24, the 87th percentile of the leave-one-country-out prediction distribution, above the out-of-sample alert rule (sensitivity 0.92, specificity 0.45) and inside the score band of the actual pre-default observations of 2018-2022 - while scoring essentially the same as Japan 2016, which never defaulted. The calibrated index saturates there and cannot separate those two histories. Median simulated debt reaches 123 percent of GDP by 2030 (IMF official: 121.8) and 142-310 percent by 2050 depending on the scenario. The exercise is a stress-equivalence benchmark, not a literal default probability; the paper's point is that debt becomes dangerous through the whole configuration around it - financing costs, liquidity, currency architecture and data coverage - not through its level.