Zi Wei Engine Validation
How we prove the Zi Wei Dou Shu engine is correct — not just claim it. Every calculation layer below is locked by an automated regression suite against an independent benchmark. Any future change that breaks a verified case blocks deployment.
Verification Layers
Tests are layered from cross-validation to derived rules. Source of truth: iztro 2.6.0, an independent MIT-licensed open-source Zi Wei Dou Shu engine — version-locked, so any benchmark drift is detected. 10,940 randomly generated charts plus 940 hand-picked edge cases form the golden corpus; the anchor chart is additionally triple-verified (iztro + manual recalculation + the YL Four-Laws theorems).
Report generated: 2026-09-12T15:19:19.721Z · Engine: ziwei_core v20260903 · Benchmark: iztro 2.6.0 (MIT) — independent open-source Zi Wei Dou Shu engine, version-locked
Public Validation Corpus
50 independently reproducible cases across 7 edge categories. These expected values were not generated by the production engine alone — every case was produced by iztro 2.6.0, an independent open-source engine, then re-verified field by field against our production engine. Anyone can take any case below, enter it into any Zi Wei calculator (ours or a competitor's), and compare.
Expected values generated by iztro 2.6.0 (MIT, independent benchmark engine) — not by the production engine alone. Cross-checked against the anchor chart triple-validation.
How the chart is calculated Zi Wei Calculator →
Run your chart — free BaZi Validation →
736-assertion sister system
How Truth Is Established
- No circular verification. We never use another astrology site's output as the expected answer — sites often copy each other's bugs. Instead: an independent open-source engine, manual recalculation, and mathematical invariants (the YL Four-Laws).
- Version-locked benchmark. The golden corpus is generated by iztro 2.6.0 exactly. If the installed benchmark version ever drifts, generation aborts — preventing silent re-baselining against a moving target.
- Edge-first sampling. The corpus over-samples where engines diverge in the wild: leap months, late-Zi hour, the Lichun–Lunar-New-Year window, gender-driven decadal direction, and Five-Element class boundaries.
- Deployment is gated. The regression suite runs before every engine release. A single failed assertion blocks the release — correctness is not negotiable.
- Public reproducibility. Every public case lists its birth data and expected output. You can verify any row in your browser right now with the Verify Live button, or type it into any other calculator and compare.