v2.0.1 - Hamel Husain Methodology

What's New in v2.0.1

• ✅ Simplified from 88 files to ~20 core files
• ✅ Binary evals (pass/fail only) per Hamel Husain methodology
• ✅ Error-analysis-first workflow
• ✅ JSONL storage (grep-friendly, version-control friendly)
• ✅ All CLI commands working: collect, view, annotate, taxonomy, eval, generate, export, report
• ✅ Fixed GitHub Pages deployment
• ✅ CI/CD passing all tests

Quick Start

npm install -g embedeval

1. Collect traces

embedeval collect traces.jsonl --output collected.jsonl

2. Annotate (error analysis)

embedeval annotate collected.jsonl --user you@example.com

3. Build taxonomy

embedeval taxonomy build --annotations annotations.jsonl

Documentation

• GitHub Pages: https://algiras.github.io/embedeval/
• README: https://github.com/Algiras/embedeval
• LLM Guide: LLM.md

Breaking Changes

Complete rewrite from v1 to v2:

• Removed: Evolution engines, complex strategies, BullMQ jobs
• Added: Simple binary evals, failure taxonomy, synthetic data generation
• Philosophy: Error analysis first, automation second