Pharmaceutical Batch Genealogy and Intervention Planning¶
Industry: Pharmaceutical Manufacturing | Role: VP Quality, Head of Manufacturing Operations | Time to value: 4 hours
The Problem¶
A deviation is found in a batch of active pharmaceutical ingredient. Every downstream batch that incorporated that material — and every batch made from those batches — is potentially affected. Identifying the full impact, assessing risk, and deciding which batches to quarantine, retest, or release is a regulatory obligation under 21 CFR Part 211 and EU GMP Annex 15. Getting it wrong means either a costly over-recall or an FDA warning letter.
The Traditional Approach¶
Quality teams trace batch genealogy through MES batch records, often across multiple systems for different manufacturing stages. A single API batch may feed 8-12 formulation batches, each producing 3-5 packaging lots. Tracing this tree manually takes 2-4 days. Risk assessment is done in spreadsheets, with quality reviewers assigning impact scores by judgment. Intervention decisions — quarantine, retest, or release — are made batch by batch, balancing risk against cost and supply continuity. The entire process is documented in Word files for regulatory submission.
With Uni¶
The notebook defines recursive campaign lineage traversal: starting from the deviated batch, Uni traces every downstream batch through campaign genealogy (NEXT_BATCH) paths, carrying risk along each path with ALONG. Risk accumulates through the genealogy — batches farther from the deviation accrue per-edge carry risk on top of process risk, so distance from the deviation drives the score. Intervention selection is cost-optimized with a dual priority via BEST BY: minimize residual risk first, then minimize cost among equally safe options. A counterfactual ASSUME scenario and an ABDUCE minimal-change search test containment, and EXPLAIN RULE produces a derivation trace for any conclusion as ready-made evidence for regulatory submission. The model is a handful of declarative Locy rules covering lineage, risk propagation, intervention costing, and optimal selection.
What You'll See¶
- Complete batch genealogy from the deviated source through every downstream campaign batch, with no manual tracing
- Risk-ranked impacted batches with scores derived from genealogy distance (hops) and per-edge carry risk
- Optimal intervention plan: quarantine, retest, or release for each batch, selected risk-first then cost-second
- Counterfactual containment (
ASSUME) and minimal-change (ABDUCE) analysis, plus a derivation trace (EXPLAIN RULE) for inclusion in regulatory deviation reports
Why It Matters¶
A pharmaceutical recall costs $10-50M on average and takes months to resolve. Reducing genealogy tracing from days to minutes and replacing judgment-based intervention with optimized, evidence-backed decisions protects both patients and the business.