Skip to content
Methodology

How SeoulOracle reads your saju

Saju isn't fortune-telling. It's a 1,500-year-old coordinate system for time that Korea has refined alongside China and Japan. SeoulOracle splits the work into two layers: deterministic computation, and an AI trained on the Korean self-pyeong (子平) tradition. This page is the transparent boundary between the two.

1. Chart computation (deterministic)

Your eight characters are an astronomical calculation — no AI involved

Your birth date and time go through a strict computation: same input, same eight characters, every time, anywhere in the world.

True Solar Time correction

Korea Standard Time (GMT+9) is a political border, not astronomical noon. We correct for the longitude offset between your birthplace and 135°E (Akashi, Japan) — Seoul (126.97°E) is roughly 32 minutes behind real solar noon, and that shifts the hour pillar.

Solar-term month boundaries

The month pillar changes on the 24 solar terms (절기), not on the 1st of the calendar month. Risechun (입춘) is the actual new year, and births in early February before it carry the prior year's pillars.

Korean self-pyeong (子平)

Different from Chinese BaZi and Japanese Shichūsuimei in how it weighs day-master strength, classifies patterns (격국), and prioritizes the Ten Gods (십신). Our engine follows the lineage formalized in 20th-century Korean clinical practice.

Unknown hour stays unknown

If you don't enter a birth hour, we leave the hour pillar blank. We do not have the AI guess. If you want to reverse-infer, our 24-question Birth Time Finder runs softmax over the 12 candidate hours using your day master and life patterns.

2. AI interpretation

What the AI does and does not decide

The computation is deterministic, but the reading is language. SeoulOracle uses Claude Opus 4.7 and GPT-5.5 with a system prompt grounded in Korean self-pyeong classics and modern clinical patterns. Here is exactly where the AI starts and stops.

What AI does NOT decide

Heavenly stems, earthly branches, five elements, Ten Gods, pattern type — these are computed by the saju engine and handed to the AI as input. The AI cannot override them.

What AI does decide

(a) Translating chart structure into clear sentences, (b) reading today's day pillar against your natal pillar in a warm Korean-elder tone, (c) drawing in Korean proverbs, seasonal references, and cultural context. Same chart, different person — the reading lands differently, and that's where the AI earns its keep.

Archetype caching

Users sharing the same archetype (day-master element × yin/yang × core pattern) share a daily reading base; only personal scores and partner data are computed individually. This is for consistency, not cost — similar charts should get similar readings.

Vision (face/palm)

Face and palm photos go through the LLM once, then are discarded immediately. No image is ever stored in our database — only the textual analysis lives in our DB for 90 days.

3. The five-element engine

How we turn eight characters into scores and prose

Eight characters need to be reduced to derived values — element balance, Ten God relationships, day-master strength — before they mean anything. Our core rules are public.

Generation and control cycles

Wood → Fire → Earth → Metal → Water → Wood (generation). Wood → Earth → Water → Fire → Metal → Wood (control). The relation between today's element and your day master's element sets your day score baseline.

Ten Gods (십신)

Each of the other seven characters is classified relative to your day master into one of ten relations — companion, rival, output, expression, direct/indirect wealth, direct/indirect officer, direct/indirect seal. The distribution of Ten Gods determines social tendencies.

Strong vs weak self

How much your day master is supported by the rest of the chart. Strong-self means lots of seal/companion support; weak-self means the chart drains the day master through output/wealth/officer. The pivot element (용신) differs depending on which side you're on.

Score formula

Overall, wealth, love, health, and career scores are weighted from (a) today × natal element relation, (b) which Ten God today's element plays for you, (c) your strong/weak status. The formula is open in apps/api/src/services/dailyReadingService.ts.

4. Limitations

What we don't promise

Saju is a mirror reflecting the texture of your time, not a blueprint of fate. A trustworthy service has to be clear about what it isn't.

Not medical, legal, or financial advice

Health language in a reading isn't a diagnosis. Wealth language isn't investment advice. Always cross-reference major life decisions with a qualified professional in that field.

Not deterministic fate

The traditional Korean line is "fate is 80%, fortune is 20%, effort is on top." Identical-saju twins diverge because every choice branches the timeline.

AI limits

LLMs are statistical language generators. Same input can yield slightly different phrasing; rarely a sentence misses the context. Treat the reading as one lens worth considering, not the final word.

Minors

Vision (face/palm) automatically rejects photos that appear to be of minors. Saju itself has no age limit, but we recommend guardian involvement below 14.

The classics we lean on

Not 'AI-generated mysticism.' Our prompts and pattern-classification logic are grounded in the following Korean and Chinese classics. We disclose sources so you can verify any claim yourself.

Qing dynasty, 18th c.

Self-Pyeong True Commentary

子平眞詮 (자평진전) · Shen Xiaozhan (沈孝瞻)

The standard text on pattern (格局) classification and pivot element (용신) selection. The most-cited primary source among Korean myeongri scholars. Our day-master strength judgments and Ten God prioritization come from here.

Ming–Qing, 14th–19th c.

Dripping Heavenly Marrow

滴天髓 (적천수) · Jingtu (京圖); annotated by Ren Tiechao (任鐵樵)

The most poetic compression of yin-yang and five-element interaction. Our reading of strong/weak self and the cold/warm/dry/wet (한난조습) lens roots here. Ren's annotated edition is the practitioner's standard.

Ming, 16th–17th c.

Gungtongbogam

窮通寶鑑 / 欄江網 · Compiled by Yu Chuntai (余春臺)

The most concrete treatment of how the season of birth (월령) shapes the day master. Our seasonal language ("a Wood day master in early spring is...") baselines from here.

Qing, 17th c.

Myeongri Concise

命理約言 (명리약언) · Chen Suan (陳素庵)

Tidied the bag of folk "spirit symbols" (신살) down to a pattern- and Ten God–centric core. Our decision to de-emphasize neopagan-style 신살 in favor of pattern + element flow tracks this book.

Song, 13th c.

Yeonhae Self-Pyeong

淵海子平 (연해자평) · Xu Dasheng (徐大升)

The effective starting point of modern myeongri. Established the day-master-centric self-pyeong method that became the bedrock of late Joseon Korean practice.

Modern

Korean 20th-century clinical lineage

韓國 命理學 臨床 · Park Jae-wan (도계), Lee Seok-yeong (자강), Baek Yeong-gwan and others

The points where Korean self-pyeong diverges from Chinese BaZi — lighter 신살 weighting, pattern hierarchy, Ten Gods prominence — come from 20th-century Korean practitioners. Our Korean cultural tone owes this lineage.

Four pledges

1

Computation is deterministic. Same input always produces the same saju — the formula is open.

2

We've drawn a clear line between AI-authored prose and engine-computed structure.

3

Photos are never stored. All vision analysis is stateless; only the text result lives 90 days.

4

Every reading reminds you we are not medical, legal, or financial advisors.