FengshuiCalculator audit

Rule Provenance Register 規則出處登記 ruleset fengshui-default-2026.1

Every rule the kernel computes, registered with the classical text it descends from, the honest attribution status of that text, and the verification that pins it in code. A citation says where an algorithm comes from — never that the source is right.

Machine-readable: rules-provenance.json · narration contract for skill layers: narration-templates.json · texts collated in the Classical Texts Register (28 titles) · contradictions logged in the School Dispute Register · math audited in the accuracy report

Methodology

Flying Stars 玄空飛星

RuleSourceAttribution statusVerification in kernel
Three Cycles Nine Periods, Lichun boundary 三元九運沈氏玄空學Late-Qing posthumous compilation (author d. 1906)Period anchors + astronomical Lichun instants (HKO-verified 1901–2100)
Luoshu flying path 洛書飛泊沈氏玄空學Universal in the lineage — no competing path knownKernel unit tests + golden charts
Downward chart, three layers 下卦盤沈氏玄空學Shen lineage systematisationGolden charts (八運子山午向=雙星會向 etc.) regressed in tests
Downward/replacement bands ±4.5°/7.5° 下卦替卦分界沈氏玄空學Practitioner convention (presupposes a modern compass)24-mountains table, kernel-consistent
Replacement-star verse 坤壬乙訣青囊奧語 · 天玉經Attribution doubted (Siku verdict) — the verse is the rule; both attributed carriers are 存疑Replacement charts recomputed + regressed; adjudication in output
Six chart patterns 六格局(閉合子集)沈氏玄空學Closed subset by design — the tradition carries moreEmpty badge = no closed pattern matched (never guessed)

Eight Mansions 八宅

RuleSourceAttribution statusVerification in kernel
Ming Gua formula, Lichun year boundary 命卦公式八宅明鏡Tang imprint doubted — content is Qing-era practice; formula in modern standardised form; two five-yellow conventions labelled (default 寄艮)Formula↔kernel assertion on all 200 year pages
Method A — Dayou-nian verse 大遊年歌八宅明鏡Eight Mansions transmissionCross-checked vs Method B, 8 palaces × 8 trigrams, exhaustive
Method B — six-line transformation 六爻變陽宅三要 (趙九峰, 4 卷)Attribution secureA ≡ B proven exhaustively in-kernel (8/8, no divergence; any future divergence would be surfaced, not hidden)

Date Selection 擇日

RuleSourceAttribution statusVerification in kernel
Officers keyed to the solar-term month 月建基準協紀辨方書 (imperial, 1739, 36 卷)Official Qing compilation — secureMonth branch computed astronomically from solar terms (the ±7.4-day lunar-month trap asserted)
Twelve Day Officers 建除十二神協紀辨方書Official Qing compilationGolden-month regression; shown for every almanac day
Yellow/Black path 黃黑道十二神協紀辨方書 卷七Official Qing compilationBoth systems computed side by side; 「萬不能與建除相合」 quoted — we take no side
Six Officers 六曜協紀辨方書 (recorded among transmitted officers)Transmitted system (flourished as Rokuyō)Uses lunar month/day — the one officer family keyed to the lunar calendar; kernel requires explicit lunar inputs, never guesses
Purple-White, year & month 紫白年白月白協紀辨方書Official Qing compilationCycled from solar-term boundaries
Purple-White, day & hour 日白時白協紀辨方書 (literature diverges)—FAIL-CLOSED: kernel returns an explicit not-closed error rather than a number — fewer numbers, no invented ones

FAQ

What is this register?

Every computational rule in the kernel — the flying-star construction, the Eight Mansions star assignments, the date-selection officers — listed with the classical text it descends from, the honest attribution status of that text, and the verification the kernel applies (cross-method exhaustion, golden charts, or explicit fail-closed).

Why do some sources carry a doubted-attribution flag?

Because tradition attributes them to authors the evidence does not support — the Tang imprint on the Qing-era Bazhai Mingjing, or the Siku cataloguers' verdict on the Qingnang Aoyu and Tianyu Jing. Our rule: the verse is the rule; the book attribution is recorded exactly as our Classical Texts Register assesses it — flagged, never laundered into false authority.

What does fail-closed mean?

Where the literature carries materially different rules and we cannot defend a closure — day and hour purple-white stars being the standing example — the kernel refuses to output a number and returns an explicit error instead. The rule is registered here as not closed.

Is a citation an endorsement?

No. A citation says where an algorithm comes from, not that the source is right. Where two transmitted systems contradict each other, we compute both, quote the classical admission that they cannot be reconciled, and take no side — the School Dispute Register logs those cases.

For entertainment and educational purposes only — this is a calculation tool for a cultural tradition, not predictions, guarantees, or advice.