What are the primary scoring dimensions in Tableu?

Readings are scored from 1-5 on personalization, tarot coherence, tone, safety, and overall quality based on a defined rubric.

Where is the evaluation data stored?

Evaluation results are stored in a Cloudflare D1 database within the eval_metrics table, capturing scores, flags, and redacted payloads for analysis.

How does the system detect hallucinations?

It uses pre-compiled patterns for domain-specific entities (like tarot cards) and context-aware filtering to ensure mentioned items actually exist in the source data.

Can I block low-quality readings from reaching users?

Yes, by enabling the EVAL_GATE_ENABLED flag, the system can block readings that trigger safety flags or fall below specific tone and safety thresholds.

Tableu Evaluation System

Name: Tableu Evaluation System
Author: henryperkins

byhenryperkins

0•

分析与监控

Monitors, troubleshoots, and calibrates automated quality metrics and safety evaluations for AI-generated tarot readings.

The Tableu Evaluation System skill provides a comprehensive toolkit for managing AI-generated content quality within the Tableu tarot framework. It enables developers to analyze narrative metrics like card coverage and hallucination detection, manage multi-dimensional scoring across personalization, coherence, tone, and safety, and configure automated quality gates to block low-quality outputs. By providing deep integration with Cloudflare D1 storage and Workers AI, it facilitates regression detection, safety auditing, and performance optimization for LLM-driven applications using sophisticated scoring rubrics and structural validation.

主要功能

01Sophisticated hallucination detection for specific domain entities

02Configurable quality gates to block unsafe or low-coherence responses

030 GitHub stars

04Automated 5-dimension quality scoring for AI-generated content

05Integrated alert system for performance regressions and safety spikes

06Real-time narrative metrics tracking including card coverage and structural validity

使用场景

01Troubleshooting why specific AI readings received low coherence or safety scores

02Calibrating evaluation thresholds and narrative metric constraints for content generation

03Auditing historical quality data and safety flags stored in D1 databases

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add henryperkins/tarot tableu-evaluation

For use in Claude.ai and ChatGPT

主要功能

01Sophisticated hallucination detection for specific domain entities

02Configurable quality gates to block unsafe or low-coherence responses

030 GitHub stars

04Automated 5-dimension quality scoring for AI-generated content

05Integrated alert system for performance regressions and safety spikes

06Real-time narrative metrics tracking including card coverage and structural validity

使用场景

01Troubleshooting why specific AI readings received low coherence or safety scores

02Calibrating evaluation thresholds and narrative metric constraints for content generation

03Auditing historical quality data and safety flags stored in D1 databases