v0.4: Stochastic Foundation

[fall-development-rob]

Summary

• Integrates multi-asset correlated return samplers (LogNormal, Student-T copula, Bootstrap), AR(1) inflation, and Gompertz/cohort longevity models into the Monte Carlo loop
• Attaches institutional risk metrics (VaR/CVaR/Sortino/drawdown/P10-P50-P90 trajectories) to MCResult when runs >= 200 and horizon >= 10 years
• Forces stochastic features off in sensitivity and backtest modules per ADR-027/031 to prevent double-counting
• Preserves byte-identical MC output for v0.3-shape scenarios (no multi-asset, Flat inflation, Fixed longevity)

Changes

• src/monte-carlo.ts: Integrated stochastic samplers; conditional trial-seed derivation to preserve legacy RNG sequence; added risk_metrics, inflation_fan_chart, lifespan_distribution to MCResult
• src/sensitivity.ts: Force-disables stochastic features on the base scenario before running the tornado sweep
• src/backtest.ts: Force-disables stochastic features for historical windows
• src/index.ts: Barrel exports for all v0.4 sampler builders, defaults, and types
• package.json: Version 0.3.0 -> 0.4.0; added "type": "module" and test script
• tsconfig.json: Added rewriteRelativeImportExtensions for proper ESM output
• scripts/fix-esm-extensions.mjs: Post-build script adding .js extensions to dist imports
• src/tests/smoke.test.mjs: 46 new assertions (68 total) covering sampler determinism, Cholesky, AR(1) mean-reversion, Gompertz median, CVaR invariants, and v0.3 backwards compat

Test plan

• All 68 smoke tests pass (22 existing + 46 new)
• npm run build succeeds
• npm run lint succeeds
• v0.3-shape MC produces byte-identical terminal values (fixture-locked)
• Integration test with app layer consuming new MCResult fields

🤖 Generated with claude-flow