Validation methodology

QuantWave's primary credibility claim is correctness: one mathematical implementation (Next<T>) powers batch Polars expressions, streaming structs, and Python bindings. This page documents how that claim is tested — every count below is machine-generated, not hand-written.

Compare vs TA-Lib · Benchmarks (measured only) · Contributing

Coverage snapshot

Last updated: 2026-07-19T06:54:45Z (UTC)

Metric	Count	Source
Native indicators (`*_METADATA`)	221	`metadata_export.json`
Metadata entries with gold file reference	217	`metadata_export.json`
Gold-standard JSON fixtures (on disk)	28	`quantwave-core/tests/gold_standard/`
Python streaming gold parity cases	26	`tests/python/gold_parity_registry.py`
Python gold parity deferred (HMM)	2	regime fixtures — separate suite
Rust `#[test]` functions (core)	583	`rg '#[test]' quantwave-core`
Rust `#[test]` functions (polars)	45	`rg '#[test]' quantwave-polars`
Rust `#[test]` functions (backtest)	92	`rg '#[test]' quantwave-backtest`
Rust tests (total, 3 crates)	720	sum of above
`proptest!` blocks (core)	160	`rg 'proptest!\{' quantwave-core`
`check_batch_streaming_parity` call sites	4	indicator modules
TA-Lib parity test functions	134	`test_*_talib_parity.rs`
Indicators with proptest parity (CI-enforced)	214	`check_indicator_parity_coverage.py`
Reviewed parity exemptions	7	`parity_exemptions.toml`

Regenerate: python scripts/collect_validation_stats.py && python scripts/render_validation_docs.py

Gold-standard vectors

Industry reference outputs are stored as JSON in quantwave-core/tests/gold_standard/. Rust indicator modules load these fixtures in unit tests; the Python package runs streaming parity against the same files via tests/python/test_gold_parity.py.

Provenance varies by indicator:

TA-Lib / talib-rs reference runs for classic indicators
Ehlers papers and Trader's Tips HTML references (see per-indicator formula_source in metadata)
Hand-verified small vectors for edge-case warmup behavior

On-disk fixtures

Each file below lives in quantwave-core/tests/gold_standard/ and is consumed by Rust unit tests and/or the Python gold parity registry.

Fixture
`alma_9_085_6.json`
`atr_ts_14_25.json`
`cycle_trend_analytics.json`
`donchian_5.json`
`ehlers_autocorrelation.json`
`ehlers_loops.json`
`ehlers_stochastic.json`
`frac_diff.json`
`fractals.json`
`griffiths_spectrum.json`
`heikin_ashi.json`
`hma_14.json`
`hmm_gaussian_2state.json`
`hmm_lambda_2state.json`
`ichimoku.json`
`keltner_20_20_15.json`
`mad.json`
`mesa_stochastic.json`
`oc_price_rsi.json`
`pivot_points.json`
`rsih.json`
`sma_5.json`
`supertrend_10_3.json`
`tema_14.json`
`ttm_squeeze_20_2_15.json`
`vortex_14.json`
`voss_predictor.json`
`wavetrend_10_21_4.json`

Streaming vs batch parity (Rust)

Every streaming indicator is encouraged to use check_batch_streaming_parity (see quantwave-core/src/test_utils.rs): for random input sequences, collecting indicator.next(x) must match the batch/plugin path within tolerance. This is the property that matters for live trading — your research DataFrame and your bar-by-bar feed must not diverge.

proptest! blocks across quantwave-core exercise this property with randomized inputs and warmup-aware comparisons.

TA-Lib cross-check

Dedicated integration tests in quantwave-core/tests/test_all_talib_parity.rs and test_missing_talib_parity.rs compare QuantWave streaming output against talib-rs batch functions for the mapped subset. This is a separate layer from gold JSON fixtures and catches drift in classic TA paths.

Python cross-language parity

The unified PyPI wheel bundles the PyO3 (abi3) core + backtest + Polars plugins extensions. Python gold parity tests assert that streaming classes match the same JSON vectors as Rust — catching FFI field swaps, warmup mishandling, or ABI mismatches before release.

CI runs tests/python/test_gold_parity.py on Linux and macOS after maturin develop (see .github/workflows/ci.yml).

Release invariants

No wheel reaches PyPI without:

Tag verification — scripts/verify_wheel_tags.py on every built wheel (abi3 tag must match extension naming)
Install smoke — Python 3.9 / 3.11 / 3.12 / 3.13 import + RSI batch/stream parity
OIDC publish — pypa/gh-action-pypi-publish via trusted publisher (release.yml); no long-lived API token

Enforced by scripts/check_release_invariants.py in CI. See GitHub Actions README.

Known exemptions

Area	Status	Reason
HMM gold fixtures (`hmm_*_2state.json`)	Deferred in Python parity	Multi-field regime outputs — separate test suite in `quantwave-core/tests/hmm_*_gold.rs`
Performance marketing numbers	Removed / harness-only	See benchmarks; unmeasured claims are blocked by `check_benchmark_claims.py`
Plugin `.ta` batch path	Partial Python gold coverage	Streaming parity lands first; Polars expression parity expands per indicator

How to extend

Add or update quantwave-core/tests/gold_standard/<name>.json
Wire Rust load_gold_standard* test in the indicator module
Add a row to tests/python/gold_parity_registry.py if Python exposes the indicator
Run ./scripts/quantwave_verify.sh locally; CI regenerates this page via check_validation_docs.py