The vintage backtest — every month of history, seasoned to term

One cohort of 100 Agreements originated at every historical month with a full 60 months of subsequent history, each replayed against the actual BTC path that followed (returns rescaled to a common $60,000 entry so dollars compare across vintages). Exits follow the three-bucket framework: ordinary defaults from the lifetime curve, underwater-tiered voluntary non-performance (drawdown-scaled hazard), above-water early settlements at full make-whole. Recoveries at mark-to-market, capped at the remaining schedule — residual coin returns to the customer. All figures net of BTC Now fees. Every assumption below is yours to sensitize.

Paper

Ordinary defaults

The hardship lottery — the framework's "ordinary defaults." Flat 10%/yr ≈ 41% over five years (as spec'd — constant hazard, no front-loading). Argue it below consumer unsecured if you like: these customers self-selected by paying.

Voluntary non-performance

Underwater-only, tiered by depth vs entry (×0.5 in the money → ×3.0 beyond −70%) — the framework's price-based trigger, driven by each vintage's real path.

Prepays & recovery

Net IRR by vintage origination month

dashed line: that month's actual BTC close (log scale) — the cycle context
no data

Every vintage, decomposed

VintageEntry closeNet IRR (eff.)MOICNet gainShortfallCompletedSettledNon-perf.BTC Now take

Recorded monthly series begins Feb 2012 — with a 60-month term that yields fully seasoned vintages (extendable if earlier bars are supplied). Same engine as the workbench: integer-cent double-entry ledger, conservation-checked per run, independently re-derived (28-check audit) and cross-validated against a second implementation to ±0.3pp on the deterministic crash grid. Same seed + same assumptions = identical results.