Test infrastructure · live

Math passes the gate before deploy

Every push to main runs the full vitest suite + TypeScript type-check in GitHub Actions before the deploy pipeline picks the commit up. A red CI run means the deploy is gated. This page summarises the suite that runs.

TEST FILES

*.test.ts under src/lib/data/__tests__/

TEST CASES

208

it() / it.skipIf() declarations

MATH INVARIANTS

forward-return engine self-test → /diag/engine

STRUCTURAL

composite math sanity → /coherence

About skipped tests: Two suites — crisis-stress-2022 and permutation-test — load 5-year price bars from per-ticker JSON files in data/backtest/. Those files are gitignored to keep the repo small (~32 MB across 327 tickers), so they exist on developer machines but not in CI. Both suites use describe.skipIf(!HAS_DATA) so they cleanly skip on CI rather than cascade-fail. They run end-to-end before any release tag.

Test inventory

Module	Test cases	Source
multiple-testing	29	src/lib/data/__tests__/multiple-testing.test.ts
scoring-rules	28	src/lib/data/__tests__/scoring-rules.test.ts
forward-return-engine	18	src/lib/data/__tests__/forward-return-engine.test.ts
zscore-composite	17	src/lib/data/__tests__/zscore-composite.test.ts
kalman-dlm	16	src/lib/data/__tests__/kalman-dlm.test.ts
conformal	14	src/lib/data/__tests__/conformal.test.ts
shapley	14	src/lib/data/__tests__/shapley.test.ts
bayesian-weights	12	src/lib/data/__tests__/bayesian-weights.test.ts
bocpd	11	src/lib/data/__tests__/bocpd.test.ts
nlp-scoring	9	src/lib/data/__tests__/nlp-scoring.test.ts
copula	8	src/lib/data/__tests__/copula.test.ts
factor-keys	8	src/lib/data/__tests__/factor-keys.test.ts
knockoff-filter	7	src/lib/data/__tests__/knockoff-filter.test.ts
crisis-stress-2022	6	src/lib/data/__tests__/crisis-stress-2022.test.ts
permutation-test	5	src/lib/data/__tests__/permutation-test.test.ts
crisis-stress-2008	3	src/lib/data/__tests__/crisis-stress-2008.test.ts
crisis-stress-2020	3	src/lib/data/__tests__/crisis-stress-2020.test.ts

What they cover

factor-keys — PRIOR_WEIGHTS sum to 1, 13 factors, every weight positive and bounded.
multiple-testing — Φ(x), p-value, Benjamini-Hochberg textbook cases, MAD-based robust σ.
conformal — empirical quantile correctness, split-conformal coverage on Gaussian + fat-tail residuals.
bayesian-weights — risk-parity blender, σ=0 floor, missing-factor handling.
bocpd — change-point detection on synthetic regime shifts, regime-probability coherence.
copula — tail-dependence (Schmidt-Stadtmüller), comonotone/countermonotone/independent extremes.
kalman-dlm — predict/update steps, posterior covariance symmetry, weight normalisation.
zscore-composite — cross-section stats, ±4σ clipping, n_effective Herfindahl.
forward-return-engine — score monotonicity, multiplier caps, NaN safety on extremes.
nlp-scoring — Loughran-McDonald sentiment + Gunning Fog readability end-to-end.
scoring-rules — Brier score (Murphy 1973 decomposition), CRPS Gaussian, reliability diagram, coverage rate.
knockoff-filter — Barber-Candès 2015 FDR control under factor co-movement, reproducibility.
shapley — efficiency invariant Σφ=v(N)−v(∅), bullish/bearish direction, FACT[13] regression test.
crisis-stress-2022 — production engine on 2022-01-04 cohort, CI coverage measurement (CI: skipped).
permutation-test — overfit detector, 1000-shuffle null distribution (CI: skipped).

External links

GitHub Actions · Tests workflow — every push runs here
/diag/engine — 46 forward-return-engine invariants live
/diag/fdr — Benjamini-Hochberg multiple-testing observable
/coherence — 7 structural invariants of the composite math

Counts derived at build time from src/lib/data/__tests__/. Numbers below the table are what vitest reports on each CI run; this page recompiles daily.