Deterministic post-processing for scores

Scores like Phase 5's `growth_score` are computed in Python after the LLM returns, not asked of the model. Reproducible across runs, auditable, tunable without re-prompting.