feat: dataset-mirror — 100x faster ingest by bulk-cloning community SFT mixes

feat: 3-track training pipeline — Kaggle T4 + Lightning H200 + (Modal pending)

feat: kaggle-trainer + llm-burst x3 — TRAIN now, MAX synthesis

feat: expand-role-keywords daemon — LLM-expanded SDLC keywords

feat: stamp-and-move — central cursor service stops re-pulling row 0

feat: llm-burst-generator daemon — synthesize 7K+ training pairs/day from 8 free LLM APIs

feat: gh-actions-ticker daemon — 60s burst dispatch (10x faster than cron)

feat: /dynamic-datasets endpoint — expose discoverer catalog to external runners

feat: self-heal-watchdog + drop shards 6→4 to dodge OOM kill

fix: 2 more Mac path leaks + /selftest endpoint to catch this pattern early

fix: REMOVE Mac dependency + boost HF Space to 16 shards + parquet-direct

feat: caps 5x bump (94M\u2192293M) + parallel 4-shard bulk ingest

feat: RAG fully wired \u2014 FTS5 + vector hybrid retrieval, no train needed for now

feat: CONTINUOUS auto-orchestrate (4 parallel workers, no 20-min gaps)

feat: HF dataset agentic discoverer \u2014 never-ending mega-mix hunt

feat: central dedup store \u2014 single source of truth across ALL writers

feat: GitHub agentic crawler — maximize 4-PAT pool (20K req/h aggregate)

feat: integrate 11 new datasets + Granite + threat-intel + SRE postmortem daemons

fix: 4 small bugs found during health check

feat: full SDLC domain expert system + 5 SWE datasets + 4 models + HF Inference Providers

feat: migrate $HOME/.claude/* to $HOME/.surrogate/* (clean separation from Claude Code)

Stream uvicorn logs to stdout + use PORT env var

Add /chat endpoint for remote CLI delegation

Replace http.server heredoc with uvicorn FastAPI (robust port binding)