fix: Disable cache for dataset generation and enhance metadata logs

feat: Add Teacher Registry, Diversity Controller, and domain refactors

feat: Real Inference Upgrade & Dataset Versioning

feat: Force add review and benchmark tools

feat: Add Research Lab tools (Review CLI, HTML Dashboard, Quality Benchmarks)

feat: Force add compiler

feat: Add Dataset Analytics & Quality Scoring. Lock Architecture 1.0.

feat: Implement Synthetic Dataset V3 Expansion

feat: Implement Synthetic Dataset Factory V2

feat: Implement the Final Capstone Training Gate

feat: Implement Verified Failure Queue, Automatic Root-Cause Classification, and Training Leaderboard

feat: Implement UI Upgrade, Custom Model deployment, Tool Registry, and Memory Engine

fix: force add ignored data pipeline changes

feat: Implement the Failure Collection Data Flywheel continuous learning loop

feat: Implement Multimodal Capability Router mapping using frozen registry architecture

fix: Use SFTConfig instead of TrainingArguments to avoid TRL 0.24 pickling bugs during checkpoint save

fix: Dynamically determine resume_from_checkpoint flag based on actual directory contents to prevent ValueError on first run

fix: Apply Unsloth ChatML template to conversational dataset to fix formatting error

fix: Update Unsloth is_bfloat16_supported API usage

feat: Implement dynamic Omni-Modal capability routing and dataset schemas

feat: Pivot architecture to Colab training and Spaces inference

chore: Prepare for Hugging Face Spaces deployment

Initial upload of Indus 10T Architecture (Hive-Mind + Q-Star)