Practitioner Architecture · AI · BaZi

Modern Tools, Traditional Practice and the Rise of AI in BaZi Reading

The useful question is not whether a machine can replace a master. It is which tool is performing which task, under which conventions, with what data, evidence, review and accountability.

Automate the repeatable. Audit the interpretation. Keep accountability human.
Do not call everything AI

Four tool classes perform different jobs.

A chart calculator can be deterministic while a generative model can vary its wording and reasoning. Treating them as one category hides the exact source of an error.

01 · CALCULATOR

Calendar and chart engine

Converts verified inputs into pillars using declared calendar, timezone, solar-time and boundary rules. Best for reproducible calculation when conventions are explicit.

02 · RULE ENGINE

Encoded interpretations

Applies prewritten rules to chart features. Consistent within its rules, but only as sound as its hierarchy, exceptions and implementation.

03 · RETRIEVAL SYSTEM

Source and case lookup

Finds relevant passages, notes or documented cases. Helpful for traceability when citations and versions remain visible.

04 · GENERATIVE MODEL

Language and synthesis

Can summarise, compare and draft explanations, but may produce variable, unsupported or confidently phrased errors.

05 · PRACTITIONER

Judgment and responsibility

Defines the question, checks context, prioritises signals, communicates uncertainty, obtains consent and remains accountable.

06 · CLIENT

Goals and lived evidence

Supplies context, corrects assumptions, decides what is relevant and retains ownership of every life decision.

Capability matrix

Match the task to the right system.

TaskCalculatorRule / retrievalGenerative modelHuman responsibility
Chart calculationAppropriate when tested and conventions are declaredCan validate expected patternsShould not be trusted as the calculation engine by defaultVerify inputs, boundary cases and output
Technical lookupLimitedUseful when sources and versions are traceableUseful for orientation, not silent authorityCheck the actual source and context
Interpretive synthesisNot its jobCan expose candidate rulesCan draft competing hypothesesWeight hierarchy, exceptions and lived evidence
Client communicationNot its jobCan supply approved languageCan help structure explanationsProtect consent, proportion, privacy and agency
High-stakes adviceInsufficientInsufficientInsufficientUse relevant licensed professionals and accountable decision processes
Human-reviewed workflow

Seven gates from birth data to responsible use.

GATE 01

Consent and purpose

State who requested the work, what question is in scope, and how the data and output may be used.

GATE 02

Input verification

Check date, local time, place, timezone history, daylight-saving treatment and data uncertainty.

GATE 03

Reproducible calculation

Use a tested engine and record calendar, solar-time, Li Chun, month-boundary and Luck Pillar conventions.

GATE 04

Independent cross-check

Compare a second trusted calculation path, especially near day, month, year or timezone boundaries.

GATE 05

Multiple hypotheses

Ask the system for alternatives, counterexamples and missing information—not one polished verdict.

GATE 06

Human review

Weight structure, context, uncertainty, language, harm, referral needs and what the tool cannot know.

GATE 07

Decision and audit

Keep the client’s choice separate from the reading. Record what was decided, why and what evidence will be reviewed.

Birth data is personal data

Do not upload first and ask questions later.

A birth date, precise time, location, name, family history and consultation notes can identify a person or expose sensitive information when combined.

MINIMISE

Use the least data required

Remove names and unrelated details. Avoid uploading third-party charts without permission.

DISCLOSE

Explain the tool

Tell the client which external system is used, why, and what categories of information it receives.

CHECK

Review current terms

Understand retention, training use, access controls, deletion, regional processing and account settings before use.

SEPARATE

Keep identity away from prompts

Use coded case identifiers and store the re-identification key separately when record-keeping is necessary.

SECURE

Control records

Protect exported charts, transcripts, screenshots, prompts and practitioner notes with appropriate access and retention rules.

DECLINE

Know when not to use AI

Highly sensitive, identifiable or regulated information may require an approved private system—or no external AI tool.

Evaluation before adoption

Test the workflow—not just a few impressive answers.

CALCULATION SET

Use boundary cases

Test timezone changes, daylight saving, midnight, solar terms, Li Chun, leap conditions and known reference charts.

INTERPRETATION SET

Use blinded cases

Score whether outputs respect hierarchy, avoid invented facts, surface uncertainty and distinguish function from event.

CONSISTENCY SET

Repeat and vary prompts

Check whether small wording changes produce incompatible claims and whether the system can cite its assumptions.

HARM REVIEW

Search for dangerous language

Flag diagnosis, financial instruction, fatalism, coercion, stigma, privacy leakage and unsupported certainty.

HUMAN BASELINE

Compare real alternatives

Measure the assisted workflow against calculator-only, practitioner-only and documented reference processes.

CHANGE CONTROL

Re-test after updates

Models, prompts, knowledge bases and software versions change. A past evaluation does not guarantee current behaviour.

Stop signs

Do not use or publish an AI reading when…

NO CONVENTIONS

The chart cannot be reproduced

The system will not state how it handled time, boundaries, calendar rules or Luck Pillar direction.

NO SOURCES

Authority is fabricated

It invents classical quotations, lineages, case histories, citations or technical rules.

NO BOUNDARIES

Advice crosses professional scope

It diagnoses illness, directs investments, determines legal action or commands relationship decisions.

NO CONSENT

Someone else is being profiled

A client asks for covert analysis of an employee, partner, child, candidate or public figure.

NO UNCERTAINTY

Every claim sounds inevitable

The language converts symbolic relationships into guaranteed events, motives or moral judgments.

NO OWNER

Nobody accepts accountability

The practitioner presents the machine as the final authority and cannot explain or defend the reasoning.

Core rule: never let polished language impersonate verified calculation, documented source, competent judgment or informed consent.

Use tools for leverage—not borrowed authority.

Make calculation reproducible, interpretation reviewable, data protected, uncertainty visible and decisions accountable.

Return to the Clarity Series