Tracker: docs/FORMULAB_V1_TASK_TRACKER.md#fvl-07--predictive-performance-engine
Supervised ML (NOT an LLM) trained on real historical formulation + experiment data (FVL-05/FVL-06) to predict measurable performance. 16 tasks (FVL-07.001..016).
Acceptance criteria: explicit INSUFFICIENT_DATA eligibility gate; baseline model compared before any ML candidate; full model registry (dataset hash, feature-schema version, metrics, applicability domain); predictions carry uncertainty + applicability-domain warning; no regulatory/safety override by prediction; per-product-family models, never one universal model.
Dependencies: FVL-05.
Scope freeze rule: see docs/FORMULAB_V1_FINAL_SCOPE.md — zero LLM in this engine.
Tracker: docs/FORMULAB_V1_TASK_TRACKER.md#fvl-07--predictive-performance-engine
Supervised ML (NOT an LLM) trained on real historical formulation + experiment data (FVL-05/FVL-06) to predict measurable performance. 16 tasks (FVL-07.001..016).
Acceptance criteria: explicit INSUFFICIENT_DATA eligibility gate; baseline model compared before any ML candidate; full model registry (dataset hash, feature-schema version, metrics, applicability domain); predictions carry uncertainty + applicability-domain warning; no regulatory/safety override by prediction; per-product-family models, never one universal model.
Dependencies: FVL-05.
Scope freeze rule: see docs/FORMULAB_V1_FINAL_SCOPE.md — zero LLM in this engine.