Decoupled Framework Design for Test-Time Learning Agents
How modular framework design decouples TaskSpec environment contracts from HarnessSpec turn orchestration, inference backends, and parallel workers.
All
AI 26
Continuous Learning 26
LoRA 19
Fine-Tuning 13
Self-Training 10
SFT 10
Continual Learning 7
Catastrophic Forgetting 6
AI Agents 5
Data Selection 3
Test-Time Learning
Reproducible ML Pipelines and Recipe Hash Provenance
How 19-field row schemas, SHA-256 recipe hashes, and atomic directory exports enforce reproducible ML across multi-generation agent fine-tuning loops.