MechInterp Glossary and Constraints

About

Serves as a specialized reference for designing valid mechanistic interpretability experiments on SplatNLP’s SAE features. It bridges the gap between game mechanics and model analysis by defining ability families, AP rungs, and hardware constraints like the ReLU floor, ensuring that experimental builds are physically possible and statistically sound. This skill is essential for researchers needing to look up ability codes, verify valid gear combinations, and avoid common analytical pitfalls like weapon-gating bias or OOD token combinations.

Key Features

• Reference of ability family short codes and AP rung values
• 0 GitHub stars
• Guidelines for handling binary tokens and activation deltas
• Standardized schemas for token parsing and build validation
• Validation rules for gear-specific and main-only abilities
• Identification of experimental pitfalls like ReLU floor quantization

Use Cases

• Planning mechanistic interpretability sweeps while avoiding out-of-distribution errors
• Verifying the validity of ability combinations in experimental builds
• Understanding the relationship between AP investment and model activations