YiGram

Methodology

Three layers, in plain sight.

YiGram is a full traditional I Ching reading first and a najia-powered decision tool second. The casting, the surface text, and the technical layer are kept separate so each one can be tested, audited, and improved on its own schedule.

Layer 1 — Casting

  • Three-coin casting that preserves the real 6 / 7 / 8 / 9 probability distribution (1/8, 3/8, 3/8, 1/8).
  • The frontend never owns the cast — it calls the FastAPI service so the same engine serves the web UI and the future MCP server.

Layer 2 — Traditional reading

  • Public-domain James Legge (1882) classical translation as the authority anchor for every hexagram.
  • A modern decision interpretation written in restrained business prose, never astrology vocabulary.
  • A line-by-line walk-through covering each of the six changing positions when they arise.
  • A short editorial provenance block on every hexagram page that names the byline, the publication date, and the sources we relied on for that reading.

Layer 3 — Najia engine (beneath the surface)

Each hexagram page exposes a small “Method notes” tab with the structural metadata that applies to that hexagram — palace, generation, shi-ying positions, line branches, six-relatives. The public hexagram pages deliberately do not describe how that structure is fed to the AI interpretation. The full pipeline lives here:

  • Hexagram palace identification, najia branch assignment, shi-ying line positions, moving-line and transformed-line markers.
  • Six-relative computation against the palace element so the AI prompt receives concrete structural context rather than vague symbolism.
  • A use-spirit decision tree keyed off the question category. Defaults are documented and labelled unaudited_draft until the source-text cross-check completes.
  • The technical terms — palace, six-relatives, shi-ying — stay hidden in the default UI and only surface behind a deliberate disclosure for readers who explicitly want them.

What V1 does not include

Monthly and day strength, combinations and clashes, hidden spirits, fan-yin and fu-yin, and other school-disputed rules sit behind Sprint 5 data. We will only add them when user evaluations show the static layer is not specific enough on its own.