Weighing Tolerance Band

A weighing tolerance band is the explicit acceptance window around a §211.101 charge target. It has four parts that must all be specified, not assumed: a target (the adjusted charge value after PF / SBF / LOD / overage / counter-balance cascade), a unit (absolute grams or percent of target — never both, never ambiguous), a directionality (symmetric ±x, or asymmetric +x / −y when over-charge and under-charge carry different risks), and a routing for each side (what the kiosk does when net lands in-band, over-band, or under-band). All four are MMR data captured under §211.186 + §211.205, locked under change-control, and rendered on the BMR at execution time so the operator, the reviewer, and the inspector all see the same window the engine enforced.

The most common mistake is treating the tolerance band as a comfort margin — 'we set it to ±2% to give the operator some room'. That is the answer that gets a §211.100 + ICH Q9 finding. The defensible answer is 'we set it to +1.5% / −1.0% based on (a) the formulation's bioequivalence dossier allowing ±3% total content uniformity contribution from charge variability, (b) the validated assay's pooled RSD of 0.8%, (c) the chosen balance's capability ratio of d/tolerance = 1/15, and (d) a QRM session that scored the risk of over-charge higher than under-charge for this potent API'. The band's WIDTH is engineering; the band's JUSTIFICATION is regulatory.

Two specification mistakes account for most tolerance-band 483s: the unit-ambiguity error (the MMR says '±2' without specifying grams or percent), and the symmetric-by-default error (the MMR says ±2% when the underlying risk is asymmetric). Both are fixable at MMR design time and impossible to fix at execution time.

A defensible tolerance band is the output of a four-input synthesis, not a single number copied from a similar product. The four inputs:

1. Formulation tolerance — the total ± content uniformity / potency variability the finished-product spec allows, less the contributions from blending + compression + analytical method, leaving the residual budget available for charge variability. Comes from the development dossier (CMC, ICH Q8(R2) design space). A 5% finished-product spec with 2% blending + 1% method variability leaves at most ~2% (RSS-combined) for charge variability across all charges in the batch.
2. Analytical method variability — the validated assay's pooled %RSD across ruggedness studies. ICH Q2(R2) suitability. A method with 0.8% pooled RSD contributes ~0.8% to the total variability budget; a method with 2.5% pooled RSD contributes ~2.5% and shrinks the budget available for charge.
3. Balance capability — the chosen balance's readability d and the d ÷ tolerance ratio. A 10:1 ratio (d ≤ tolerance / 10) is the industrial baseline; tighter is better, looser is unresolvable. A 4-place analytical balance (d = 0.1 mg) supporting a ±2 mg tolerance has a 20:1 ratio — comfortable. A bench balance (d = 0.01 g) supporting a ±5 mg tolerance has a 0.5:1 ratio — unworkable.
4. ICH Q9(R1) risk assessment — formal QRM session scoring the harm + likelihood of over-charge vs under-charge, capturing the asymmetric-risk decision, the risk-control rationale for the chosen band, and the residual-risk assessment that justifies the band against the design-space budget. Documented at the MMR-approval gate, not retrofitted after the first OOS.

The band is meaningless without explicit routing for each region. The kiosk must execute a deterministic decision tree at the moment net is captured — there is no operator-discretion route that says 'just close enough'.

1. WO release: dispense engine computes the per-charge target as scaled_nominal × PF × SBF × LOD × (1 + overage) and freezes it into the WO snapshot.
2. MMR snapshot binds tolerance band cells to each charge line — unit, direction, low value, high value, top-off margin, and routing references — all under change-controlled MMR version.
3. Operator opens the kiosk charge: tolerance band displayed numerically + visually (range slider with target, low edge, high edge, top-off region highlighted).
4. Pre-flight: balance capability check (d ≤ tolerance / 10 baseline; warn at 10:1, hard-block at 4:1), min-weight check (charge ≥ min-weight × k_safety), max-load check.
5. Tare-verified weighment chain (pre-zero / tare / stability / gross / post-drift) produces the net mass.
6. Net captured: kiosk evaluates against band; routes per the decision tree (under / under-topoff / in / over).
7. On in-band: weighment.accepted_in_band fires; dispense_result written with charge, target, tolerance band, balance_id, operator_id, container_id, tare/gross/net trio, calibration_valid_until, post-tare drift, UTC timestamp; e-sig captured; witness e-sig prompted for requires-witness materials.
8. On under-band-topoff: kiosk computes top-off increment, opens top-off sub-weighment, re-judges; chain repeats until in-band or over-band.
9. On over-band or hard-under: deviation auto-opens with original-vs-target delta pre-populated; QA review required before WO advances.
10. BMR closure: every charge's net, target, tolerance, routing outcome, and any deviation reference rendered side-by-side; §211.192 reviewer judges the cumulative tolerance compliance against §211.103 yield closure.

1. Unit-ambiguous MMR — line reads '0.500 ± 0.005' with no unit; operator and reviewer interpret differently; kiosk allows release; OOS later shows charge variability outside any intended band; §211.186 + §211.188 deficiency.
2. Symmetric band on asymmetric risk — formulation can tolerate +3% but only −1% (over-potent is rejection, under-potent is remediable); MMR specifies ±2%; over-charges in 1-2% range fall in-band but should have been routed under-band-topoff or rejected.
3. Band wider than balance capability — ±5 mg specified on a balance with d = 0.01 g (capability 0.5:1); the band is mathematically unresolvable; every weighment lands 'in-band' because the balance cannot distinguish, but the actual variability is unbounded.
4. Band tighter than formulation budget — ±0.1% specified when the variability budget is 2%; operators cannot hit the band; chronic top-off rate >10%; weighment time triples; eventually band is widened reactively in change-control with 'aligned to operational reality' justification — a §211.100 + ICH Q9 finding.
5. No QRM document for band choice — band copied from similar product, no formal Q9 session, no design-space tie-back; first inspector question 'how did you arrive at this tolerance?' has no answer.
6. Kiosk allows operator override of over-band weighment — 'accept anyway' button skips deviation; audit trail shows in-band acceptance even though net exceeded band; data-integrity violation; routinely cited under MHRA DI 2018.
7. Band judged against nominal instead of post-adjustment target — operator weighs to nominal 500 mg ±5 mg; post-PF/SBF/LOD adjustment the true target is 480 mg; weighment at 502 mg lands 'in-band' against nominal but is +4.6% over the adjusted target; latent OOS contributor.
8. BMR omits routing outcome — BMR shows the net mass but not whether it was in-band / topoff / over; §211.192 reviewer cannot audit the routing; cumulative band-rate analytics are impossible.

• In-band first-pass rate — fraction of weighments that land in-band on the first weighment (no top-off, no rejection); target ≥95% for routine excipients, ≥98% for actives; sustained low rate indicates band-too-tight or balance-capability gap.
• Top-off rate — fraction of weighments requiring at least one top-off; target ≤5%; sustained high rate indicates band-too-tight, operator-training gap, or balance-capability gap.
• Top-off events per topoff'd charge — average count of top-off increments per charge that required at least one; target ≤1.2 (most top-offs land in one increment); sustained ≥2 indicates operator training + balance pre-flight gaps.
• Over-band rejection rate — fraction of weighments rejected as over-band (cannot be remediated); target ≤0.5%; non-zero is always a deviation + material-loss + investigation.
• Band-vs-capability ratio compliance — fraction of MMR tolerance bands with d ÷ tolerance ≤ 1/10 against the assigned balance; target 100% (any sub-10:1 is a control-strategy gap).
• Band-vs-formulation-budget compliance — fraction of MMR tolerance bands with documented design-space tie-back per ICH Q8; target 100% (any missing tie-back is a regulatory gap).
• Asymmetric-band utilisation — fraction of actives + high-potency materials with asymmetric bands where QRM justified asymmetry; target reflects the QRM portfolio (not all materials warrant asymmetry, but symmetric-by-default for actives is a finding).
• Operator-override attempts — count of attempted 'accept anyway' on over-band weighments; target 0 (the kiosk does not expose the override; non-zero indicates a UI regression or a config drift to be investigated).

1. MMR tolerance cells are first-class: each charge line carries tolerance_unit (g | pct), tolerance_low, tolerance_high, topoff_margin, routing_under, routing_over, asymmetric_rationale_ref. Unit + direction + routing are mandatory at MMR approval; the approval gate refuses to advance an MMR with any missing cell.
2. Engine evaluates band against the POST-adjustment target (scaled_nominal × PF × SBF × LOD × (1 + overage)), not the nominal — band-judged-against-nominal is a closed code path, not a config flag.
3. Balance pre-flight enforces d ÷ tolerance ≤ 1/10 baseline as a soft warning (operator + supervisor see the capability ratio) and ≤ 1/4 as a hard block. The kiosk refuses to start a weighment on a balance whose capability cannot resolve the band.
4. Kiosk widget renders the band visually (range slider, target marker, low/high edges, top-off region highlighted, current reading tracked live). The operator sees the band, not just the number.
5. Routing decision tree is deterministic and audit-trailed: every weighment commits with the routing_outcome field (under_rejected | under_topoff | in_band | over_rejected) and the routing reference (deviation_id or topoff_chain_id).
6. Top-off sub-weighments are chained: each top-off carries a topoff_chain_id linking back to the original under-band weighment; the BMR renders the full chain, not just the final net.
7. Over-band rejection auto-opens a deviation with original-vs-target delta + material-disposition workflow (return-to-source vs scrap) pre-populated.
8. MMR change-control captures every band edit: width change, unit change, direction change, routing change. Quarterly band-rate aggregates feed the ICH Q10 §3.2.5 product review; chronic band-rate drift triggers a formal control-strategy revisit, not a band-widening shortcut.
9. Operator-override is not exposed in the kiosk; the only path past an over-band weighment is a Part 11 e-sig'd deviation with QA approval, fully audit-trailed.
10. SPC + analytics: per-product per-balance per-operator in-band-first-pass rate + top-off rate + over-band rejection rate trend on the Quality dashboard; control-limit breaches page the production supervisor + open an investigation candidate before yield drifts.