SYSTEM WORKFLOW

From raw narrative generation to deployed inference models.

The `teacher` module uses LLMs to generate initial narrative scenarios and their corresponding TKS equations.

python scripts/run_teacher.py

Applies a series of inversions and anti-attractor syntheses to create a large, diverse, and balanced training corpus.

python scripts/generate_augmented_data.py

The new Rust compiler pipeline processes TKS source through Lexing, Parsing, Desugaring, Resolution, Inference, Lowering, and Codegen.

tksc build main.tks --emit bc

Trains a model on the augmented dataset to accurately encode natural language into TKS expressions.

python scripts/train_with_augmented.py

Comprehensive validation using canonical rules, validator pass-rate computation, and per-augmentation-type breakdown.

python scripts/phase6_eval.py

The model is used for inference for single or bulk processing, or served as a web service.

python scripts/run_inference.py