A 500,000-row synthetic instruction dataset focused on:
self-writing code generation
self-debugging via patches + tests
self-morphing refactors (typed modernization)
self code enhancements (optimization)
self-evolving agentic planning (offline-first, safety gated)
Plain JSONL shards in data/ (no gzip).
Each row:
id
instruction
response
metadata (genesis curriculum + tags)
