FateryBench

Best AI

43.9%

Fatery (Gemini 3 Flash + prompt engineering + pre-computed BaZi chart)

Human Top 3

60.0%

HKJFMA 2024 competition 3rd place. AI still 16% behind human experts.

Fatery Effect

+13.8%

Prompt engineering + chart data lifts Gemini 3 Flash from 30.1% to 43.9%.

Best Baseline

38.3%

GPT-5.4 without any prompt engineering. All baselines cluster 30-38%.

Leaderboard

Scored on 196 HKJFMA competition questions (2022-2025). Celebrity cases excluded due to data contamination. Human reference: HKJFMA 2024 3rd place = 60% (source: arXiv:2510.23337). Random baseline = 25%.

Rank	Model	Provider	Approach	Accuracy	Score

The Fatery Effect

Same model (Gemini 3 Flash), same questions. Fatery adds a professional system prompt and pre-computed BaZi chart data — four pillars and Da Yun cycles calculated by a deterministic algorithm.

Results Explorer

Browse every question and see how each model answered. Click a row to expand the full case.

ID	Source	Category	Question	Answer

Baseline vs. Fatery

Baseline

Simple prompt — the model must calculate everything from scratch.

          System: 你是一位八字分析师

          命主信息：男命 1974年4月28日…

          问题：此命1996年发生何事？

          选项：A / B / C / D

          请先排出八字命盘，分析后给出答案。

Fatery

Production prompt from Fatery.me — structured analysis with pre-computed chart.

          System: 命运精算师 — 4-step protocol

          ├ STEP 1  命盘分析（日主·强弱·用神·忌神）

          ├ STEP 2  宏观周期弧（大运）

          ├ STEP 3  针对性分析（冲合刑害）

          └ STEP 4  选项评估

          + Pre-computed 四柱八字 via lunar-python

          + Pre-computed 大运 (8 decade cycles)

Dataset

196 multiple-choice questions from four years of HKJFMA (Hong Kong Junior Feng Shui Masters Association) competitions across 12 categories.