Parent–Child Batch

Every batch produces output material that becomes input to other batches. The parent–child link is the explicit, persistent record of that flow:

• Parent — the producing batch (intermediate, sub-assembly, bulk pack).
• Child — the consuming batch (next-step intermediate, finished product, packaging lot).
• Quantity — how much of the parent went into the child (with units).
• Timestamp — when the consumption occurred.
• Equipment — which unit performed the transfer.
• Operator/signature — who confirmed the consumption (where SOPs require).

The links are graph edges; the batches are nodes. The resulting genealogy graph supports traversal in both directions: backward from child to parent ('what made this?') and forward from parent to children ('where did this go?').

• One-to-one — parent batch wholly consumed by one child (rare in practice).
• One-to-many (split) — parent material divided across multiple children (common: bulk to multiple package SKUs).
• Many-to-one (merge) — multiple parents combined into one child (common: blending sub-lots into a final blend).
• Many-to-many — multiple parents split across multiple children (campaign manufacturing).
• Self-referential (reprocess/rework) — original batch is parent of the reprocessed batch with the same product code.

• Backward traversal — given a child batch, recursively walk parent edges to enumerate every raw material lot consumed, every intermediate, every operator and every piece of equipment involved.
• Forward traversal — given a parent batch, walk child edges to enumerate every finished good, every shipment, every distribution point.
• Quantity reconciliation — the sum of parent contribution into children plus losses should equal the parent's output quantity; deviations flag genealogy gaps.
• Depth bounded by recipe complexity — most consumer-product chains are 3–6 levels deep; API to finished drug can be 10+.
• Time bounded by retention — links live as long as the longest-retained batch in the chain.

Parent–child links are what makes recalls scopeable in hours rather than days. FDA recall classifications expect 'identified scope' in the first 24 hours; FSMA 204 imposes 24-hour KDE export for affected foods. The graph traversal must produce:

• Every finished-good lot traceable to the suspect batch (forward).
• Every customer who received those lots (forward, via shipment records).
• Every other batch that shared the same raw material lot (sideways via shared parents).
• Every piece of equipment used in the suspect chain (for inspection and CIP review).

A mock recall is the only honest test of whether the parent–child graph is recall-ready.

• Capture at the moment of consumption — barcode-scan or weigh-event captures the parent lot ID at the transfer; never typed later.
• Reject typed lot IDs if a barcode is available — typing introduces drift.
• Quantity captured with unit and tolerance — '120.5 kg ± 0.5' not '120 kg'.
• Partial consumption captured cumulatively — same parent referenced by multiple consumption events across the child's life.
• Material substitutions explicit — when an approved alternate lot is used, capture both the planned and actual lot.
• Rework explicitly modelled — the rework batch's parent edge points at the original, with rework rationale recorded.

• Pharma — API lot → granulation sub-lot → blend lot → compression sub-lots → coating sub-lots → packaging SKU lots.
• Biopharma — cell bank vial → seed train batches → production fermentation batch → harvest pool → DS lots → DP fill lots → individual vials.
• Food — raw ingredient lots → mix lot → fill lot → case → pallet, with FSMA 204 KDE capture at each CTE.
• Cosmetics — bulk emulsion lot → multiple fill SKU lots (tube/jar/bottle), each with its own packaging genealogy.
• Chemicals — feedstock lots → reaction batch → distillation cuts → blended product lot, with each cut individually traceable.

• Lot IDs typed at consumption — drift between planned and actual, broken genealogy.
• Bulk consumption recorded as one event when actual transfers happened over hours — quantity reconciliation fails.
• Rework not modelled as parent–child — appears as fresh batch; original quality issues lost.
• Sub-lot identity lost when material is binned — 'unknown lot' nodes break recall queries.
• Equipment not captured per parent–child transfer — CIP residue paths cannot be reconstructed.
• Cross-site transfers handled by manual reconciliation — links break at the site boundary.
• Mock recalls run only when audited — undetected genealogy gaps surprise during real incidents.