In-Process Reconciliation

In-Process Reconciliation is the mid-batch closure step where, at the end of a defined process stage, the mass / count / volume balance for that stage is reconciled against the scaled theoretical: dispensed-in minus measured-out minus authorised-loss equals Δ; |Δ| must be inside the stage's validated tolerance before the batch is allowed to progress to the next stage. Run after every stage that produces a measurable in/out pair — typically after dispense, after granulation, after compression, after coating, after filling, and after packaging.

It is distinct from end-of-batch yield reconciliation, which closes the whole batch in a single calculation against the final-product theoretical. End-of-batch reconciliation tells you the batch lost 2.3% somewhere. In-process reconciliation tells you the batch lost 2.1% at compression and 0.2% at coating — so investigation focuses on compression-die wear or punch sticking, not on a five-stage hunt. It is also distinct from IPC (in-process controls), which tests quality attributes (assay, hardness, dissolution, content uniformity, fill weight) at sample points. The two are complementary: IPC tests whether the batch is right; in-process reconciliation tests whether the batch is whole.

Σ in − Σ out − Σ authorised loss = Δ. |Δ| ≤ stage tolerance (declared on the MMR, derived from PPQ campaign data). Stage tolerance is typically tighter than end-of-batch tolerance — e.g. ±0.5% per stage vs ±3% end-of-batch — because losses accumulate across stages and a single stage burning the whole budget leaves nothing for the rest.

An authorised loss is a mass / count / volume that left the batch through a validated, expected pathway. Every authorised-loss category must have evidence in the BMR — the loss is not authorised just because the operator says it happened. Six categories cover most regulated manufacture:

1. IPC samples taken — sample mass recorded contemporaneously at the sample point; the IPC result and the sample mass close out together in the BMR.
2. Retain samples taken — per 21 CFR 211.170 retain quantity (twice the amount required for full testing); mass recorded; transferred to the retain inventory with chain-of-custody.
3. Validated equipment hold-up — quantified during PPQ for each piece of equipment in the train (mixer dead-legs, granulator hold-up, tablet-press dust pockets, coater overspray-on-walls, filler dead-volume, packaging conveyor losses). The validated figure is on the MMR; using anything other than the validated figure is a deviation.
4. Validated dust extraction — solid-dose plants typically have local dust-extraction ducting that pulls a measurable mass out of the process; the validated rate is captured per stage. If the actual extraction rate diverges from validated, the dust-extraction system itself is in a deviation state and the affected batches need investigation.
5. Line clearance discards — material left in the equipment, on the line, or in transit that is discarded before the next batch starts; mass / count recorded; discarded in a controlled manner.
6. Operator-damaged or weight-rejected units — finished or semi-finished units rejected by automated weight-check or by operator inspection; rejected units are counted (or weighed), evidence is captured (reject reason code), and rejected units go to a controlled-destruction or recovery pathway.

Stage-gate reconciliation switches unit-of-measure at the format-change point. Bulk stages (dispense, granulation, compression-bulk-in, coating-bulk-in) reconcile in mass. Unit stages (after compression, after coating-tablet, after filling, packaging) reconcile in count. Liquid stages (compounding, holding-tank, filling-line-in) reconcile in volume (with a density cross-check to mass for tank closure).

The format-change point is the most-cited recurring 483 in this area. At compression, mass-in (granulation output kg) converts to count-out (number of tablets) via average tablet weight; at filling, count-in converts to mass-out (kg packaged) via the same conversion. The conversion must be against the validated average tablet weight from the IPC weight-uniformity testing, not against the nominal tablet weight from the MMR. The validated average tablet weight typically differs from nominal by 0.5–2% — using nominal makes the mass / count conversion always slightly wrong, and a 1.5% conversion error becomes a 1.5% phantom loss at every format change.

1. No stage gates, only end-of-batch — the only closure is at packaging-complete; a 2.3% batch loss is recorded with no stage attribution; investigation outcome is 'process loss within acceptable range' for years; the underlying compression-die-wear pattern stays invisible until a CAPA-driven re-validation surfaces it.
2. IPC samples not declared as authorised loss — operator takes a 50 g IPC sample but the mass doesn't appear in the stage reconciliation; the 50 g shows up as a Δ; the deviation triggers; QA writes it off as 'IPC sample, not investigated'; the data-integrity habit is established and real losses get hidden in the same noise.
3. Equipment hold-up not validated — MMR uses an estimated mixer hold-up figure rather than a PPQ-measured figure; stage reconciliation passes by 0.3 kg but the actual hold-up is 0.7 kg larger; phantom +0.4 kg appears at the next stage and disappears again at the stage after that as the figures balance out over time.
4. Count vs mass confusion at unit-format change — compression closes in mass, filling opens in count; the conversion uses nominal tablet weight instead of validated IPC-measured weight; a systematic 1.5% phantom loss appears at every format change.
5. Line-clearance discards untracked — the floor sweep at end-of-batch is discarded without being weighed; the discard is an unaccounted-for negative on the next stage's reconciliation; batches with longer line clearances show worse reconciliation, but the trend is invisible because the discards aren't measured.
6. Reconciliation done after the stage closes (retroactive close) — operator finishes the stage, hands over to the next shift, then comes back hours later to close the reconciliation when the QA team asks; the contemporaneity requirement of ALCOA+ is breached; the closing data is no longer a contemporaneous record.

• Stage-gate coverage % — fraction of process stages on each MMR that have a defined in-process reconciliation gate (target 100% for stages with measurable in/out).
• Contemporaneous-close % — fraction of stage-gate reconciliations closed within the validated time window of stage completion (target 100%; ALCOA+ contemporaneity).
• Authorised-loss-evidenced % — fraction of declared authorised losses with the required evidence in the BMR (sample mass, retain transfer, equipment-hold-up validated figure, line-clearance discard log) (target 100%).
• Format-change conversion accuracy — fraction of mass↔count conversions using the contemporaneous validated IPC-measured average weight, not nominal (target 100%).
• Stage-gate Δ-in-tolerance % — fraction of stage gates closing within validated tolerance (target ≥98%; trending below alerts to upstream drift).
• Stage-gate breach investigation closure time — median days from breach to investigation closure (target ≤14 days; per ICH Q9(R1) timely investigation principle).
• End-of-batch surprise rate — fraction of batches whose end-of-batch yield reconciliation surfaces a loss not predicted by stage-gate trend (target 0%; surprises mean stage-gate coverage or tolerances are wrong).

1. MMR schema: stages[] array on the MMR carries name, in_uom (kg / units / L), out_uom, validated_tolerance, validated_hold_up, validated_dust_extraction, validated_overspray, and the conversion-factor source (e.g. 'IPC weight-uniformity result, mean of n=20').
2. WO release: stages[] is copied into work_orders.mmr_snapshot as part of the YA-release scaling — every stage's hold-up / extraction / tolerance is scaled to the chosen batch size where the validated relationship is linear, or held to discrete validated values where it isn't.
3. Stage-gate UI: at the end of each stage, the kiosk opens the reconciliation form — in mass / out mass / authorised losses by category with evidence attached, computed Δ, validated tolerance, and a stage-gate e-sig (operator + QA reviewer where required by SOP).
4. Hard-block on breach: |Δ| > tolerance blocks the next stage from opening, auto-opens a deviation, pages the QA reviewer, and surfaces the breach on the batch dashboard with red status.
5. Format-change conversion: at every stage where in_uom ≠ out_uom (e.g. kg-in / units-out at compression-close), V5 pulls the contemporaneous IPC-measured conversion factor (mean of the IPC weight-uniformity test, with n and sd captured) and uses it for the conversion; the conversion factor + source ID are written into the BMR.
6. Authorised-loss evidence: every authorised-loss row requires the evidence type its category requires (sample mass field for IPC samples; retain-transfer doc number for retains; validated-figure ID for hold-up / extraction; line-clearance log entry for line clearance).
7. Contemporaneity gate: the stage-gate reconciliation form cannot be opened more than the validated time window (typically 30 minutes) after the stage's measured-out event without an explicit late-close deviation; this is the ALCOA+ contemporaneity enforcement.
8. Trending dashboard: stage-gate Δ values feed an SPC chart per product per stage; out-of-control points (Western Electric / Nelson rules) surface as early-warning signals to the process-engineering team before they accumulate to end-of-batch yield problems.
9. BMR rendering: each stage's reconciliation appears in the BMR as a numbered subsection — in, out, authorised losses with evidence, Δ, tolerance, e-sigs, and any deviation reference. End-of-batch yield reconciliation summarises the chain and closes against the scaled theoretical.
10. Genealogy and CPV: stage-gate Δ data feeds the upstream-to-downstream genealogy view (so a lot's contribution to each stage's loss is traceable) and the ICH Q10 §3.2.5 PQS management review (so leadership sees the per-stage loss pattern at the level it can act on — equipment, SOP, training, supplier).