Audited Divination for AI Agents with Oraclebone

Description and use

Oraclebone provides a deterministic, auditable layer for AI divination. By separating result generation (script) from interpretation (LLM), it eliminates hallucinations in symbolic structures. It supports Tarot, I Ching, Xiao Liu Ren, and Bazi, with specific bug fixes for traditional calculation errors (e.g., Xiao Liu Ren off-by-one, Bazi zodiac boundaries).

Problem

AI chatbots hallucinate divination results (e.g., non-existent I Ching hexagrams, incorrect Bazi pillars based on lunar vs. solar terms) because they generate the outcome rather than computing it. This makes the results un-auditable, non-reproducible, and symbolically invalid.

Outcome

Oraclebone provides a deterministic, auditable layer for AI divination. By separating result generation (script) from interpretation (LLM), it eliminates hallucinations in symbolic structures. It supports Tarot, I Ching, Xiao Liu Ren, and Bazi, with specific bug fixes for traditional calculation errors (e.g., Xiao Liu Ren off-by-one, Bazi zodiac boundaries).

Workflow

• Install the Oraclebone toolkit via pip or use the MCP server via uvx.
• Run a divination command (e.g., oraclebone tarot --spread three-card) to generate a deterministic result.
• The script outputs JSON containing the draw/hexagram/pillars and audit metadata (method, seed, warnings).
• Pass the JSON output to an AI agent (via MCP, CLI, or agent skills) for interpretation.
• The AI agent interprets the provided symbols within safety boundaries, refusing medical/legal/financial advice.
• Review the interpretation, which is grounded in the audited, script-generated result.

Evidence

The model never generates the divination result. It reads one.

Source

Fixed with ((month + day + hour - 3) % 6) + 1 and anchored test cases against known traditional results.

Source

Births between lichun and lunar new year produced contradictory output... The zodiac is now derived from the year pillar's earthly branch

Source