All integration notes
Use these detailed notes for version boundaries, method settings, source revisions, and environment-specific behavior. Some existing notes are in Chinese; the main guides introduce each topic in English.
Usage and experiments
- English usage and Python API · 简体中文使用指南
- Choose and run an experiment
- Automatic experiment launcher
- Snapshot evaluation reference
- Snapshot storage
- SearchQA × DeepSeek reproduction
- Documentation overview
Method implementations
| Method | Guide |
|---|---|
| OPRO / GEPA / EvoPrompt / SkillOpt | Algorithm index and YAML examples |
| ACE | Playbook, modes, and parameters |
| Evo-Harness | Native skill loading and task streams |
| AHE | Edits, audits, and manifest |
| ADAS | Design archive and reflections |
| DarwinX | Branches, recombination, and admission |
| Meta-Harness | File history and candidate search |
| ModularRSI | Modules, integration, and evolution data |
| RRSI | Selection rules, budgets, and costs |
Benchmarks and environments
- All 30 benchmark configurations
- Harness / benchmark boundary
- BBEH and SuperGPQA
- AIME
- Coding and SpreadsheetBench 2
- Terminal-Bench · DeepSWE
- Web research services
- τ²-bench · ALFWorld · WorkBench
Dataset and evolution sources
Harness implementation details
- Agent adapter and optional container execution
- Codex setup and capability limits
- Codex pinned baseline and Python port
- Codex task mechanisms
- Codex migration and evolution plan
Development history
These are historical implementation notes and plans. Use the main guides and module documentation for the current interface.