root c3faa570ca docs: add core-training.md - three approaches for training the heuristic scheduler ...

- RL approach: MDP formulation, PPO, curriculum learning
- LLM-driven approach: compact LLM fine-tuned as scheduler
- Neuro-symbolic approach (recommended): symbolic rules (70%) + small neural policy (30%) + arbiter
- Expression verifier training for Core's internal verification component
- Comprehensive training pipeline combining all approaches
- Evaluation metrics for Core performance

2026-06-01 10:00:16 +08:00

..

architecture-overview.md

docs: fix architecture boundaries - routing=unique input, LLM=unique output, memory=autonomous

2026-06-01 09:38:33 +08:00

core-training.md

docs: add core-training.md - three approaches for training the heuristic scheduler

2026-06-01 10:00:16 +08:00

llm-subnet.md

docs: fix architecture boundaries - routing=unique input, LLM=unique output, memory=autonomous

2026-06-01 09:38:33 +08:00

logic-trace-subnet.md

fix: correct architecture docs - tensor protocol, heuristic scheduling, LLM dual output, memory update mechanism

2026-06-01 09:20:14 +08:00

memory-subnet.md

docs: fix architecture boundaries - routing=unique input, LLM=unique output, memory=autonomous

2026-06-01 09:38:33 +08:00

routing-compression-subnet.md

docs: fix architecture boundaries - routing=unique input, LLM=unique output, memory=autonomous

2026-06-01 09:38:33 +08:00