For Practitioners 執業者的驗證頁

You audit everything else you use professionally. This page gathers, in one place, every piece of evidence about how — and how well — this engine computes. No marketing: numbers, links, and the methods that produced them.

Everything below is regenerated by a 27-layer acceptance suite that must pass before any change ships.

The one-sentence version

Every chart this engine produces is reproducible from public information — engine version + ruleset + a SHA-256 input digest you can replay in your browser, against our API, or with your own tooling. When we disagree with other engines, the disagreement is named, quantified and attributed to a convention. When we were wrong, it's in a public errata.

The proof points

Claim	Evidence	Verify it yourself
Day & hour pillars never diverge from any reference implementation	~157,000 samples: lunar-javascript 9,100 · tyme4ts 55,360 · sxtwl 102,360 (two depths, incl. 1986–1991 DST windows and pre-1970 strata) — zero divergence; the invariant fails the build if violated	Accuracy report §7
Year/month divergences are all named, never unclassified	Every divergence attributed: boundary-instant precision (second-level, vs HK Observatory) or convention granularity (D-07)	Report §7 · Dispute register D-07
Solar terms & lunar months match the golden source	ytliu/DE441 independent reconstruction: 4,800 terms at 99% within ±2 minutes; 2,473 lunar months with 200-year leap-month zero divergence (incl. the 2033 edge case)	Report §5 · Term tables
Calendar matches the Hong Kong Observatory for 1901–2100	Full-range machine cross-check; 99.88% of month starts bit-identical, every divergence attributed (midnight-margin physics + pre-1929 LMT)	In-browser verifier · Report
Geomagnetic declination uses official models, routed by survey date	WMM2025 matches all NOAA official vectors; IGRF-14 (1900–2025) differentially tested against a community implementation of the same official table (120 cases)	Declination entry
Where we differ from other engines is visible, live	Three engines (this kernel, lunar-javascript, tyme4ts) compute any chart side by side in your browser; divergences tinted and attributed	Comparison tool
When we were wrong, you can read about it	Public errata register — every past error, its magnitude and its fix	Errata

A 5-minute audit you can run right now

Open the comparison tool with a birth moment you know cold — including the awkward ones (立春 day, 23:00, a leap-month date). See whether the divergence pattern matches your understanding.
Pick any chart from your case files, enter it in the four-pillar calculator, and copy the fingerprint (engine version + ruleset + digest).
Replay the digest — the format is public: in your browser via the verifier, or POST /v1/replay with the original request.
Check the errata and dispute register for the conventions that matter to your practice.

Weekly almanac digest 黃曆週報

One email a week: the coming seven days' day officers, solar-term boundaries worth knowing, and one classical-text excerpt. Double opt-in — the subscription only takes effect when you open the confirmation link; unsubscribing deletes your data entirely. No birth data, ever — this list is email-only.

For practices that compute at volume

The public API gives you the same kernel over HTTPS — calendar, four pillars under all four convention combinations, solar terms, Kua numbers, flying-star charts — free, keyless, fair-use limited, with Pro subscribers lifting the limits via their existing license key. AI agents get the same catalogue through MCP. Every response is provenance-stamped, so your client records stay independently verifiable.

Link & cite

If this material is useful in your teaching, writing or client work, link to the specific tool or proof — deep links are stable: /compare/, /audit/report-v1.html, /disputes/, /errata/, /docs/fif/. If you find a divergence this engine's differentials don't explain, that is exactly what the errata process is for — published practitioners get credited.

FAQ

What makes this engine different from other charting tools?

It shows its work: every result carries the engine version, ruleset and a SHA-256 input digest that anyone can replay; the day/hour pillars hold a zero-divergence invariant across roughly 157,000 cross-implementation samples; year/month divergences are attributed by convention in a public dispute register; and known past errors are kept in a public errata instead of being silently patched.

Can I check your numbers against the tools I already use?

Yes — the chart comparison tool runs this kernel, lunar-javascript and tyme4ts side by side in your browser on any birth moment you choose, and explains each divergence. The server-side differentials additionally cover sxtwl and an independent DE441 reconstruction across roughly 290,000 samples.

Can my practice use the API?

Yes — the public API is free and keyless with fair-use limits, and Pro subscribers raise the limits with their license key. Every response carries provenance and a replayable digest, so your records stay verifiable independently of this service.

Are you claiming your conventions are the correct ones?

No. Where lineages differ (year conventions, day boundaries, chart kinds), all alternatives are computed and shown side by side, and the disagreements are documented in the dispute register. The claim is that the computation is exact and reproducible — not that any school's interpretation is authoritative.

All pages

Four-pillar chart · Engine comparison · Flying star · Kua number · 24 mountains · Zodiac by year · True solar time by city · Almanac · Glossary · Classical texts · Methodology · Accuracy report · Disputes · Errata · API & docs