CHANGELOG.md

Changelog

All notable changes follow Keep a Changelog format. This project adheres to Semantic Versioning.

[1.3.0] — 2026-07-04

Maintenance + correctness sweep: traditional 时辰 boundaries, CI runner deadline, qimen school-note, subcommand coverage, README refresh.

Fixed

• huangli 时辰 boundaries (correctness) — shichen_detail used even clock blocks (00-02, 02-04 …) that straddle two traditional 时辰, mislabeling the second half of every block. Now uses the classical odd-start convention (子 23-01, 丑 01-03 … 亥 21-23) with a shichen label per block; each block's 干支 branch now provably equals its 时辰 (regression-tested). NOTE: hour_range values in output changed — hence the minor version bump.

Added

• Qimen 三元 school note: determine_ju documents the 简化日数法 vs 拆补置闰法 divergence (±1 元 near 节气 edges) in both the script docstring and references/06-qimen.md — honest approximation, no unfounded claims.
• tests/test_subcommands.py (+6, suite 1007 → 1013; coverage 82.5 → 84.8%): hand-verified yijing numbers golden (3/5 → 火风鼎50, 变 火水未济), yijing text / meihua name determinism, xiaoliuren solar golden + 子时 boundary, huangli traditional-boundary regression lock.

Changed

• CI actions bumped for the GitHub Node20 runner removal (2026-09-16): checkout v4→v5, setup-python v5→v6, upload-artifact v4→v5.
• READMEs (中/EN): badges + metrics refreshed (1013 tests / 85% coverage / 15 engines / 7-check harness), new feature bullets (解读纪律 CI-lock, optional quantum entropy, exploration tool), methods table gains the exploration row.

[1.2.0] — 2026-07-04

Interpretive-discipline release: classical sources become the binding rule.

Added

• SKILL.md「解读纪律 (Interpretive Discipline) — 古籍为纲」 — BaZi judgments must anchor in the five classics with an explicit precedence order: 《子平真诠》(格局) →《滴天髓》(强弱气势) →《穷通宝鉴》(调候, already shipped as assets/tiaohou.json) →《三命通会》(神煞杂断) →《渊海子平》(十神六亲). Hard rules: 凡古籍无据者不妄断 (label folk-lore/school views as such or stay silent); 禁止套话和迎合 (no platitudes, no flattery-softened verdicts); only the strongest-evidence, most-verifiable conclusions (chart-anchored, 应期 falsifiable, classic-citable); 学理/民俗 layered; explicit conflict-resolution order (调候 vs 格局, classics vs modern schools).
• check_interpretive_discipline in evals/run_checks.py (harness now 7 checks) — CI-locks the discipline text in SKILL.md and the classics anchor in agents/openai.yaml so it cannot silently regress.
• references/01-bazi.md header now carries the binding 论断依据 note.
• agents/openai.yaml default_prompt extended with the same discipline for OpenAI-runtime consumers.

Changed

• evals/run_checks.py: ruff-clean (import order, capture_output).

[1.1.9] — 2026-05-31

Randonautica-inspired exploration tool (honest, no pseudoscience).

Added

• scripts/explore_cast.py — 今日随机寻访点: QRNG (reuses entropy.py) → uniform random points in a radius → dependency-free grid-density anomaly (attractor / void / power / blindspot, all clamped inside the circular radius) → bearing + distance + 16-point compass, cross-referenced with today's 黄历 吉神方位 (财神/喜神/福神). Carries a safety block and an explicit disclaimer: it is a randomized walk prompt, NOT a prediction and NOT a mind-matter-interaction (MMI) device — intention is recorded, never biases the entropy. SKILL.md routes 随机寻访/探索 to it.
• tests/test_explore.py (+12, suite 989 → 1001) — within-radius for all 4 modes, seed determinism, geometry (haversine/bearing/compass), input validation, 黄历 alignment, safety/disclaimer presence.

Note

Borrowed only the legitimate tech from Randonautica (QRNG + spatial density anomaly + intention UX + safety). The "intention biases quantum RNG / z-score = psi" MMI claim is explicitly rejected, consistent with this project's stance that divination efficacy is not a physically-measurable quantity.

[1.1.8] — 2026-05-31

Optional quantum entropy source for divination casts.

Added

• scripts/entropy.py — pluggable cast entropy: seed (deterministic), system (OS CSPRNG, default), or quantum (QuantumRandom, physical randomness from ANU quantum-vacuum noise, gracefully degrading to os.urandom with a degraded flag if the source is unreachable).
• --entropy {system,quantum} on yijing_cast, liuyao_cast, tarot_draw; output carries an honest entropy provenance block.
• tests/test_entropy.py (+12, suite 977 → 989) — source selection, forced-offline degrade, getrandbits/shuffle/choice correctness, and script wiring. Network-free (the quantum path is tested via the fallback).

Note

The quantum source is offered as a physically-true randomness option only. It does not make a reading more accurate — hexagram/card outcomes are uniform regardless of entropy source, and divination accuracy has no physical dependence on where the bits come from. Output always labels the source so the distinction stays transparent. (Relativity/quantum mechanics cannot improve divination accuracy; the calendar layer's solar-term precision already uses relativistic time scales via lunar_python's VSOP87 port, to < 1 s.)

[1.1.6] — 2026-05-31

Independent-verification + quality-gate release. Cross-checks the calendar engine against a second codebase and wires lint + coverage gates into CI.

Added

• Differential tests vs sxtwl (tests/test_differential.py) — an INDEPENDENT engine (C++ port of 寿星天文历). Cross-checks 日柱 over a 447-date grid (1920-2080) and 年/月柱 on all non-节气 days; both engines agree. This closes the "self-snapshot" gap (the rest of the suite validated bazi_calc against the very library it wraps). Also asserts the 立春 year-pillar switch is time-aware (flips at the exact instant, verified more precise than sxtwl's date-level API).
• lint + coverage gates — ruff (config in pyproject.toml) and subprocess-tracked coverage with fail_under = 80 (real total 82%; subprocess tracking via COVERAGE_PROCESS_START since most tests drive the CLIs out-of-process). Both wired into CI.
• Tests for lunar_convert (公历↔农历 round-trip) and zodiac info/year/taisui sub-commands (suite 94 → 977 with the differential grid).

Changed

• Cleaned all ruff findings across 16 scripts: removed unused imports/vars, deduped 4 repeated keys in the name 笔画 fallback (same-value, no behaviour change), moved module imports to top, renamed ambiguous l, added explicit zip(strict=...). ziwei_calc now surfaces true_solar_time_applied.

[1.1.5] — 2026-05-31

Bug-fix release — three silent-wrong defects found by an adversarial line-by-line audit, plus value-level test coverage to lock them.

Fixed

• 塔罗 (HIGH) — tarot_draw.load_deck gated on isinstance(deck, list), but the asset ships as {major_arcana, minor_arcana} (dict), so the curated 78-card deck never loaded and every reading silently used placeholder text ("…第N阶: 见详细解读"). Added _flatten_asset_deck; readings now carry the real per-card 正/逆位 meanings.
• 黄历 吉时/凶时 (HIGH) — classification used "时辰 has any 宜", which is true for all 12, so output was always 12 吉 / 0 凶 (meaningless). Now derived from the 时辰 黄道/黑道 (getTimeTianShenLuck): a real 吉/凶 split. Removed dead getTimes() / getTimeXun() calls.
• 紫微 真太阳时 (HIGH) — --tz/--longitude were accepted but ignored, so a user-supplied longitude produced an uncorrected chart (silent-wrong near 子时). Now wired through longitude_correction (with day roll-over) — opt-in: only applied when a non-default longitude/tz is given, so 时辰-granular charts on the default meridian are unchanged (no regression).
• 生肖 (LOW) — same-sign pairs (e.g. 鼠-鼠) were mis-flagged 三合 because da in group and db in group is true when da == db. Now requires distinct branches.

Added

• +5 tests (suite 94 → 99): value-golden assertions for 紫微 命宫/身宫/局, 奇门 阳遁8局, 大六壬 丙午日; regression locks for the tarot asset, huangli 黄黑道 吉凶 split, ziwei longitude opt-in, same-sign 三合; replaced a tautological meihua relation truthy-check with a valid-relation assertion.

[1.1.4] — 2026-05-31

Continuous integration — closes the last engineering gap found in a 4-repo competitor scan (only the off-topic Master-skill had CI; no fortune skill did).

Added

• .github/workflows/ci.yml — on every push / PR to main: install deps, run the 94-test pytest suite, the run_checks.py release harness, and build_skill.py, on Python 3.11 + 3.12; uploads the built skill zip as a CI artifact. Concurrency-cancelled, pip-cached, least-privilege permissions.
• requirements-dev.txt — dev/CI deps (runtime + pytest).
• CI / tests / license badges on both READMEs.

[1.1.3] — 2026-05-31

Packaging + bilingual-install pass — one-command distributable for Claude Code, Claude.ai upload, and OpenAI/other runtimes.

Added

• scripts/build_skill.py — self-validating, deterministic packager that emits dist/chinese-fortune-v<version>.zip. Whitelists runtime files (SKILL.md, references/, scripts/ runtime, assets/, agents/, READMEs, LICENSE), excludes all dev/test cruft (tests/, evals/, pycache, .bak, _competitors, the builder itself), nests under chinese-fortune/, and aborts on bad frontmatter / over-long description / non-compiling script.
• tests/test_build.py (+4, suite now 94) — asserts SKILL.md at package root, runtime files present, ZERO dev-cruft leakage, and that a freshly extracted package runs standalone.

Changed

• README.md / README.zh.md: replaced the single cp -r step with a 3-target Install table (Claude Code unzip · Claude.ai upload · OpenAI adapter) plus the build_skill.py one-liner. Stays concise.

[1.1.2] — 2026-05-31

Test-coverage + agent-hardening pass, informed by a 2026 market scan of best-in-class 命理 engines (cantian-ai/bazi-mcp, 6tail/lunar-python & tyme4ts, sxwnl, SylarLong/iztro) and academic evals (Celebrity-50, BaziQA).

Added

• Engine test coverage (tests/test_engines.py, +15 tests, suite now 90): contract + determinism tests for the 8 previously-untested engines (周易, 梅花, 六爻, 小六壬, 生肖合婚, 奇门, 大六壬, 黄历) plus 紫微 structure, and a table-free 五鼠遁 hour-stem invariant verified across 5 charts. Seeded casts asserted reproducible; 六冲/三合 compatibility asserted by score.

Changed

• agents/openai.yaml default_prompt hardened: restates script-first computation, references/ grounding, disclaimer, and the red-line refusals — so the OpenAI adapter carries the safety layer even before SKILL.md loads.

Notes

• Precision re-classified as already-solved. Market scan confirmed lunar_python's 节气 engine is a port of sxwnl's ShouXingUtil (VSOP87, mean 节气 error < 1s) — i.e. already at the top-tier ephemeris bar. The earlier "no high-precision ephemeris" concern was a false deduction; the only remaining numeric approximation (Spencer EOT, ±20s) is negligible against 2-hour 时辰 buckets.
• Honest ceiling. Remaining depth gaps (per-method golden corpus, iztro 紫微 cross-check, LLM-judge interpretation eval) require validated external datasets and are intentionally not fabricated. Divination truth is not scientifically validatable; engine correctness is — and that is what the test suite now locks.

[1.1.1] — 2026-05-26

Engineering-hardening pass (no reading-logic changes). Closes blockers from a CTO-grade code audit; raises correctness, determinism, and test rigor.

Fixed

• 真太阳时 day roll-over (correctness) — utils.longitude_correction clamped near-midnight times to the same day, corrupting the 日柱 (day pillar) for western/eastern longitudes. Now returns (day_offset, hour, minute); bazi_calc.py and qimen_cast.py apply the offset to the date before deriving pillars.
• Operator precedence in lunar_convert._serialize 节气 lookup (A or B and C) → explicit null-guarded branch.
• Non-deterministic output — bazi_calc.py 流年 used datetime.now().year; added --as-of-year for reproducible output.
• Silent wrong strokes — name_analyze.py defaulted unknown chars to 8 strokes; now merges FALLBACK_BIHUA under the asset (fixes missing common chars e.g. 涵=12), adds a reliable flag, and a --strict mode that refuses estimation.
• 00-foundations.md 天干相克 label "5克" → "10克 (阳干5 + 阴干5)".
• evals/run_checks.py printed ok per check before a later check failed (misleading); now collects results and prints a PASS/FAIL summary with correct exit code; stopped false-flagging gitignored __pycache__ (only TRACKED cache fails).

Added

• Input validation in bazi_calc.py (month/day/hour/minute/year bounds) returning structured errors before touching lunar_python.
• pytest suite (tests/, 72 tests) — golden values for 十神/五行/60甲子/真太阳时, midnight roll-over regression, bazi end-to-end snapshots, determinism, input validation, name reliability.
• Machine assertions for all 12 eval scenarios (evals.json) + check_eval_assertions and check_unit_tests wired into run_checks.py (deterministic substrate now verified, not just described).
• Pinned lunar_python>=1.4.4,<2.0.

Removed

• scripts/bazi_geju.py + scripts/ziwei_patterns.py (1666 LOC) — unused (zero imports) and divergent from the inline 格局/pattern logic in bazi_calc.py/ziwei_calc.py. Consolidated to a single source of truth. The inline engines remain the active, tested implementations.

Known deferred (non-blocking)

• Shared constant tables (旬空/六冲/季节五行) still duplicated across a few cast scripts (identical values, low risk). The real divergence hazard (differing 格局 thresholds) was in the removed dead modules.

[1.1.0] — 2026-05-16

Major depth + safety upgrade after deep competitive code analysis of top 6 GitHub rivals (jinchenma94/bazi-skill 1420⭐, hhszzzz/taibu 156⭐, Horace-Maxwell/horosa-skill 136⭐, china-testing/bazi 1316⭐, Renhuai123/ziwei-doushu 563⭐, cantian-ai/bazi-mcp 373⭐). All algorithms re-derived from classical public-domain sources (《穷通宝鉴》《滴天髓》《紫微斗数全书》《奇门遁甲秘籍大全》《六壬大全》).

Added

New methods with computational scripts

• scripts/qimen_cast.py (833 lines) — 奇门遁甲 时家盘: 局数自动判定 (节气+三元), 三奇六仪 地盘/天盘排布, 八门九星八神飞布, 8 种格局检出 (三诈/天遁/地遁/人遁/青龙返首/飞鸟跌穴/击刑/入墓)
• scripts/liuren_cast.py (647 lines) — 大六壬 时课: 月将加时, 四课, 三传 (5法: 贼克/比用/遥克/伏吟/反吟), 12 天将昼夜布盘, 用神 keyword routing

Pattern detection modules

• scripts/bazi_geju.py (746 lines) — 八字格局自动判定: 特殊格 (从财/从杀/从儿/从势/化气/一行得气/两气成象) + 10 正格 + 破/纯/救应判定
• scripts/ziwei_patterns.py (920 lines) — 紫微 24 格局检测: 6 上格 + 8 中格 + 4 副格 + 6 凶格

New assets

• assets/tiaohou.json — 《穷通宝鉴》调候用神 120 entries (10 干 × 12 月支), 含季节、五行状态、primary/secondary 用神、寒燥分

Changed

SKILL.md upgrades

• Frontmatter description appended activation directive ("即使只提到 ... 也主动调用")
• New 9-step Information Collection Protocol with AskUserQuestion / plain text dispatch
• New Edge Cases dispatch table (10 scenarios: 时辰未知/节气交界/夜子时/闰月/海外/双胞胎/收养 etc.)
• New Closed-Loop Calibration step in Workflow (3-5 已发生 events for user verification)
• New Required Output Fields section enforcing 用神/格局/真太阳时 surface in every BaZi reading

Script improvements

• scripts/utils.py — added Equation of Time (Spencer formula) to longitude_correction(); new true_solar_time_info() returns full breakdown with EOT contribution (±16 min seasonal variation)
• scripts/bazi_calc.py (448 → 1003 lines) — wired all 35 神煞 (vs 9 before) via SHENSHA_CATEGORY dispatch; added 用神/喜神/忌神 selection (扶抑+调候 综合); 月支本气×3/中气×1.5/余气×0.8 weighted 五行; 干支互动 detection (天干五合/地支六合/三合/三会/六冲/六害/三刑); 自动判格 (delegated to bazi_geju); 真太阳时校正 surfaced in output
• scripts/ziwei_calc.py (488 → 1041 lines) — added 6 吉星 (左辅右弼文昌文曲天魁天钺), 6 煞曜 (擎羊陀罗火星铃星地空地劫), 9 杂曜 (天马红鸾天喜孤辰寡宿天哭天虚龙池凤阁), 命主/身主 by 年支, 斗君, 自化 detection per 宫干, 大限四化, 流年四化 via --liu-year, 借宫 for empty palaces, 14 主星亮度 (庙旺平陷), 24-pattern 格局 detection, fixed 大限顺逆 bug for 阴男阳女

Validation

• evals/run_checks.py — added qimen_cast + liuren_cast to check_core_scripts test matrix; all 4 checks pass

Stats vs 1.0.0

• Files: 62 (was 51, +11)
• Markdown: 12,627 lines
• Python: 9,148 lines (was 3,825, +138%)
• Total: ~21,775 lines

License attribution

All algorithms re-derived from public-domain classical Chinese metaphysics sources. No code copied from AGPL or proprietary repos. Inspiration credit to competitive landscape audit (jinchenma94/bazi-skill UX patterns; hhszzzz/taibu architecture concepts; Horace-Maxwell/horosa-skill envelope patterns; ziwei-doushu pattern catalog structure) — interfaces and design patterns only, no source.

---

[1.0.0] — 2026-05-16

Added — initial public release

Core skill

• SKILL.md — 123-line router with frontmatter trigger description (covers 25+ Chinese & English trigger keywords)
• agents/openai.yaml — OpenAI-compatible runtime metadata for cross-platform invocation

References (23 files, ~11,540 lines)

• 00-foundations.md — Yin-Yang, 5 elements, 10 stems, 12 branches, 60 Jiazi, 8 trigrams, 24 solar terms, time pillars, 10 Gods, 12 life stages
• 01-bazi.md — Four Pillars: chart construction, day-master strength, 10 Gods, shensha, patterns, luck cycles, annual interpretation, 6 family relations, health, three worked examples
• 02-ziwei.md — Zi Wei Dou Shu: 12 palaces, chart steps, 14 main stars + assistants, 4 transformations, 三方四正, 大限, classic patterns, two worked examples
• 03-yijing.md — I-Ching: 三易, 十翼, 阴阳爻, 64 hex formation, 6 casting methods, changing lines, 互/综/错/变卦
• 04-liuyao.md — Liu Yao: 8 palaces, 世应, 六亲, 六神, 纳甲 full table, 用神, 10-step casting procedure
• 05-meihua.md — Mei Hua Yi Shu: 7 casting methods, 体/用 core, 5 generation/control relations, 外应, 10 application categories
• 06-qimen.md — Qi Men Dun Jia: 3 boards, 9 palaces, 3 wonders, 6 instruments, 8 gates, 9 stars, 8 gods, layout procedure, 12+ patterns
• 07-daliuren.md — Da Liu Ren: 月将, 4 lessons, 3 transmissions (9 methods), 12 generals, 9 schools
• 08-fengshui.md — Form school + 八宅 + 玄空飞星 + 三元九运 + 24 mountains + 形煞 + internal layout + modern reinterpretation
• 09-mianxiang.md — Face: 3 zones, 5 features, 12 palaces, 5 face shapes, moles, lines, complexion, modern thin-slicing parallel
• 10-shouxiang.md — Palm: 5 main lines, 8 trigrams in hand, 7+5 hand types, finger joints, nails, life-line timing
• 11-cezi.md — Glyphomancy: 8 techniques (拆/添/减/反/谐音/字象/字意/笔画), 5 case studies, character-element mapping
• 12-huangli.md — Almanac: 12 jianchu, 28 lunar mansions, 10 event categories, 三煞, 太岁, 彭祖百忌, full daily structure
• 13-qiming.md — Naming: 5-grid analysis (天/人/地/外/总), 81 numerology (full table), 三才, BaZi-based supplementation, company naming
• 14-hehun.md — Marriage compatibility: 3 methods, 12×12 zodiac matrix, 6 BaZi axes, modern meaning
• 15-jiemeng.md — Dream interpretation: 6 dream types, traditional + Freud/Jung, ~80 common symbols across 10 categories
• 16-shengxiao.md — Chinese zodiac: 12 detailed entries, 三合/六合/相冲/相刑/相害, 60 Jiazi pairings, 太岁 (本命/冲/刑/害/破)
• 17-xingzuo.md — Western astrology: 12 signs, 4 elements × 3 modes, planets, houses, aspects, 12×12 compatibility
• 18-tarot.md — 78 cards (22 major + 56 minor by suit), 7 spreads, reading procedure, vs I-Ching comparison
• 19-shensha.md — Auspicious & inauspicious shensha: 16 + 19 entries with full 起法 (calculation rules)
• 20-disclaimer.md — Red lines, ethical boundaries, crisis-handoff template, language safeguards
• 21-extended-methods.md — Coverage matrix for 14 rare methods (Tai Yi, Tie Ban, Cheng Gu, Hetu-Luoshu, Seven Politics, Yan Qin, Xuan Kong Da Gua, Dou Shou, Ling Qian, Bei Jiao, Zhuge, bird/omen, etc.)
• 64hex-full.md — All 64 hexagrams: classical 卦辞 + 大象 + 384 lines (王弼通行本) + 用九/用六 + 白话 summary

Scripts (12 files, ~3,825 lines)

• bazi_calc.py — Full BaZi: 4 pillars, hidden stems, 10 Gods per pillar, 5-element count (surface + hidden), nayin, shensha (9 categories), 大运 cycles, 流年
• ziwei_calc.py — Zi Wei: 命/身宫, 五行局, 紫微星position, 14 main stars, 12 palaces, 三方四正, 大限, year-干 transformations
• yijing_cast.py — I-Ching: 4 casting methods (coins/numbers/time/text), main/nuclear/changed hex, full classical text via assets/64hex.json
• liuyao_cast.py — Liu Yao: extends yijing with 京房八宫, 世应, 纳甲 (per-trigram), 六亲, 六神, 旺相休囚, 月破/日破/旬空
• meihua_cast.py — Mei Hua: time / numbers / name casting, 体/用 with 生克比和, seasonal strength
• xiaoliuren_cast.py — Xiao Liu Ren quick cast (no dependencies): 6-palace cycle, lunar/solar input
• huangli_query.py — Daily almanac: 12 jianchu, 28 mansions, 宜/忌, 吉时, directional gods, 彭祖百忌, 胎神, 冲煞
• lunar_convert.py — Solar ↔ lunar with jieqi, ganzhi, zodiac, 28-xiu
• name_analyze.py — Naming: 5-grid + 81 numerology + 三才, with 2,594-char Kangxi stroke table
• zodiac_compat.py — Zodiac info, 12×12 compatibility (1-10 score), year-zodiac lookup, Tai Sui check
• tarot_draw.py — Tarot: 5 spreads (one/three/celtic/relationship/daily), full 78-card deck, seedable
• utils.py — Shared constants: 干/支/五行/八卦/藏干, 十神 computation, 五虎遁/五鼠遁, longitude correction, UTF-8 JSON printing, graceful lunar_python guard

Assets (11 JSON files, 211 KB)

• ganzhi.json — 10 stems + 12 branches + 60 Jiazi + nayin + 5 he + 4 san-he + 4 san-hui + 6 chong + 4 xing + 6 hai
• wuxing.json — 5 elements with full property map + 旺相休囚死 by season
• bagua.json — 8 trigrams with binary, nature, family, body, animal, directions
• 64hex.json — 64 hexagrams: judgment + image + 6 lines each (+ 用九/用六 for 乾/坤)
• ziwei_stars.json — 14 main + 6 auspicious + 6 malefic stars, 10 year-stem transformations, 5 wuxing-ju
• shensha.json — 16 auspicious + 19 inauspicious shensha with 起法 tables, 6 旬空, 三合五行 group
• 24jieqi.json — 24 solar terms with 节/气 marker + BaZi month mapping
• tarot78.json — 22 major + 56 minor arcana (upright + reversed meanings)
• jiemeng.json — ~80 dream symbols (traditional + modern psychology)
• name_bihua.json — 2,594 Kangxi-dictionary stroke counts
• name_shuli.json — Full 81-numerology table

Validation

• evals/evals.json — 12 test cases covering all major methods
• evals/run_checks.py — 4-check release harness: frontmatter strict (name + description only, ≤1024 chars, 9 mandatory triggers); all scripts emit valid non-error JSON; all routed references exist; no TODO/TBD/placeholder/pycache leftover

Documentation

• README.md — English
• README.zh.md — Simplified Chinese
• LICENSE — MIT + cultural-content disclaimer
• CONTRIBUTING.md — Bilingual contribution guide
• CHANGELOG.md — This file

Safety

• Hard-coded red lines refuse: death prediction, medical/legal/financial advice, curses, third-party blame, fee demands, product recommendations
• Crisis-handoff template for self-harm / acute distress signals
• Disclaimer auto-emitted on every chart-based reading

Known limits

• ziwei_calc.py covers 命/身宫 + 14 main stars; assistant stars (副星) and 自化 / 流年 飞星 marked as scope for v1.1
• 奇门 / 大六壬 / 太乙 lack computation scripts (reference-only for now)
• jiemeng.json at 80 entries (target 500+ for v1.1)
• assets/64hex.json covers 王弼通行本 only; alternative transmissions not included