271 ·GitHub
What is this?
Yueyuan (跃渊) is a research-style BaZi (Chinese Four-Pillars) analysis Skill: it converts a birth date and time into a four-pillar chart with a built-in deterministic engine, then reads it through classical Ziping methodology while flagging evidence levels and uncertainty instead of giving flat predictions.
Yueyuan (跃渊) is a research-style BaZi (Chinese Four-Pillars) analysis Skill: it converts a birth date and time into a four-pillar chart with a built-in deterministic engine, then reads it through classical Ziping methodology while flagging evidence levels and uncertainty instead of giving flat predictions.Is this a chatbot app, or something an AI assistant runs as a tool?It's packaged as a 'Skill' - a Python engine (baziengine.py, built on lunar-python) that handles the exact chart math, paired with rule files and methodology docs that an AI assistant follows for interpretation. It is not a standalone app with its own UI.
It's packaged as a 'Skill' - a Python engine (baziengine.py, built on lunar-python) that handles the exact chart math, paired with rule files and methodology docs that an AI assistant follows for interpretation. It is not a standalone app with its own UI.How do I actually use it - do I need to compute my chart first?No, you can just give the skill your birth date, time, and gender and the built-in engine charts it. You can also feed it an already-computed four-pillar chart (e.g. a screenshot or text from a charting app) and it will cross-check both.
No, you can just give the skill your birth date, time, and gender and the built-in engine charts it. You can also feed it an already-computed four-pillar chart (e.g. a screenshot or text from a charting app) and it will cross-check both.Does it rely on any external library for the chart math?Yes, the deterministic engine baziengine.py is built on the lunar-python library and handles all calendar/pillar conversion; it's a Python script, and no other runtime is mentioned.
Yes, the deterministic engine baziengine.py is built on the lunar-python library and handles all calendar/pillar conversion; it's a Python script, and no other runtime is mentioned.How reliable are its predictions, especially about timing of events?Static structure judgments (strength, pattern, favorable element) tested around 96-100% accuracy, but predicting exact event years is still weak at roughly 50% - it only gives candidate windows and decision criteria, not guaranteed years.
Static structure judgments (strength, pattern, favorable element) tested around 96-100% accuracy, but predicting exact event years is still weak at roughly 50% - it only gives candidate windows and decision criteria, not guaranteed years.Who is this actually built for?It's meant for people who want a serious, falsifiable reading of their own chart and are willing to verify conclusions against real life experience, not for anyone wanting a flat fortune-telling verdict.
It's meant for people who want a serious, falsifiable reading of their own chart and are willing to verify conclusions against real life experience, not for anyone wanting a flat fortune-telling verdict.What makes it different from an AI just guessing at a BaZi reading?It pre-registers its predictions before seeing your reported life experience, then scores them as hit/partial/miss/unknown against what actually happened, and every conclusion is tagged with an evidence level and a rule tier (STRUCTURAL > CORE > SCHOOL > HEURISTIC > SUPPLEMENTAL) so the reasoning is traceable rather than just asserted.
It pre-registers its predictions before seeing your reported life experience, then scores them as hit/partial/miss/unknown against what actually happened, and every conclusion is tagged with an evidence level and a rule tier (STRUCTURAL > CORE > SCHOOL > HEURISTIC > SUPPLEMENTAL) so the reasoning is traceable rather than just asserted.What safety limits (SAFE red lines) does it enforce?It refuses fatalistic verdicts, medical diagnoses, and claims like predicting widowhood, divorce, or terminal illness; anything touching health is framed only as a traditional-metaphysics tendency, never a diagnosis.
It refuses fatalistic verdicts, medical diagnoses, and claims like predicting widowhood, divorce, or terminal illness; anything touching health is framed only as a traditional-metaphysics tendency, never a diagnosis.What does a finished analysis look like?By default it produces an HTML report color-coded by the five elements, covering the core verdict, derivation chain, luck-cycle/year windows, and domain readings like marriage or career; if HTML isn't supported it falls back to an equivalent Markdown report.
By default it produces an HTML report color-coded by the five elements, covering the core verdict, derivation chain, luck-cycle/year windows, and domain readings like marriage or career; if HTML isn't supported it falls back to an equivalent Markdown report.How was its accuracy actually measured?Against an external v3 diagnostic test set, static judgments like strength and pattern scored about 96-100% and SAFE-rule compliance was 100%, but the overall 85% figure from that test is explicitly called an internal diagnostic number, not a marketing accuracy claim.
Against an external v3 diagnostic test set, static judgments like strength and pattern scored about 96-100% and SAFE-rule compliance was 100%, but the overall 85% figure from that test is explicitly called an internal diagnostic number, not a marketing accuracy claim.What are similar projects?ex-skill, Evidence-Bound-Press-Conference-Revision-Skill, bigpeng-hot-gzh, SkillCorpus, skill, sun-style-writing, hypoarena, codex-claude-academic-skills
ex-skill, Evidence-Bound-Press-Conference-Revision-Skill, bigpeng-hot-gzh, SkillCorpus, skill, sun-style-writing, hypoarena, codex-claude-academic-skillsWhich topics is it in?rules, skill-md, evaluation, claude-skills, skill, testing, benchmark, agent-skills, research, chinese, tool, cli
rules, skill-md, evaluation, claude-skills, skill, testing, benchmark, agent-skills, research, chinese, tool, cli