The engineering decision
Agree a batch record schema before production. Retain individual results, configuration identity, exceptions and a route from shipped serials back to their evidence.
Have these inputs ready
- Pack serial and batch structure
- Required tests and acceptance limits
- Cell/lot and configuration mapping
- Deviation, rework and release process
Link each shipped serial to its build records
Scroll horizontally to view the full diagram →
Ask for identity, measurements and disposition
A certificate stating that a batch passed does not show the spread of the measurements or whether a unit was reworked. Agree which individual records are retained, which are delivered with each batch and which are available on request under the project data agreement.
Keep units, limits, method revision, timestamp and instrument identity with measured values. A resistance number without SOC, temperature or measurement timing may not be comparable across batches. A BMS firmware version without its configuration and parameter set leaves the actual protection behavior ambiguous.
| Record group | Minimum useful fields | Decision supported |
|---|---|---|
| Identity | Pack serial, batch, model and BOM revision | What construction was shipped |
| Cell input | Model/lot, screening method and group assignment | Incoming variation and containment |
| Software | BMS hardware, firmware and configuration checksum | Actual operating/protection behavior |
| Process | Relevant joining/torque records and fixture revision | Controlled assembly |
| EOL | Measured values, limits, method and result | Unit acceptance |
| Exceptions | NCR, rework, deviation and approval | Disposition of nonconforming work |
| Release | Approver, date, quantity and serial range | Authorized shipment |
A batch average hides the tail
In a hypothetical 100-pack batch, suppose 99 units have a measured path resistance of 4 mΩ and one has 10 mΩ. The batch average is only 4.06 mΩ. Yet at 100 A the outlier would dissipate 100 W in that measured path compared with 40 W for a 4 mΩ unit, assuming resistance remains fixed.
The example does not establish an acceptance threshold. It shows why an average-only report cannot support individual acceptance or isolate a high-loss unit. Retain individual values and connect each result to a serial.
| Population | Measured resistance | Loss at 100 A |
|---|---|---|
| 99 units | 4 mΩ each | 40 W each |
| 1 outlier | 10 mΩ | 100 W |
| Batch average | 4.06 mΩ | Does not identify the outlier |
Make the trace work in both directions
From a returned serial, retrieve the cell lots, released design, configuration and test history. From a suspect cell lot or programming revision, identify all affected packs and shipment batches. This reverse lookup determines the scope of containment.
Preserve raw records when a test is repeated. Mark invalid measurements with a reason, retain rework history, and distinguish first-pass yield from final accepted yield. Otherwise a repeated-test workflow can conceal a deteriorating process.
Use stable identifiers and controlled access. Define retention time, file formats, ownership and backup responsibilities in the project agreement. A screenshot of a dashboard is not a substitute for retrievable structured records.
Review trends and exceptions before release
Compare distributions and chronological trends with the qualified process baseline. Investigate shifts even when all readings remain inside limits; a drift may precede a defect. Keep process control limits separate from engineering specification limits.
Qualify sampling plans for the failure modes they address. Some evidence is destructive or periodic, while other checks apply to every unit. State this distinction in the control plan instead of presenting sampled coverage as 100% inspection.
Measurement & validation procedure
| Test / review | Record | Acceptance condition |
|---|---|---|
| Trace a sample serial | Linked identity, input, process and test records | All links resolve to controlled evidence |
| Run reverse containment | Query from suspect lot/configuration to serial list | Affected shipped and held units are identifiable |
| Review batch release | Outliers, trends, rework and deviations | Exceptions have an approved disposition |
Continue this engineering task
Related engineering guidance
Prepare the evidence for review
- Current QC report or proposed schema
- Serial/lot mapping example
- Test methods and acceptance limits
- Record retention and deviation requirements
Technical sources & claim support
References support the stated engineering principles. Worked examples use their explicitly stated assumptions.
- ISO 9001:2026 — Quality management systems: requirements and official publication status ↗
Quality management systems; consistency of products and services
Quality-system context for documented information, process control and traceable production records.
- NIST/SEMATECH: Assessing Process Stability ↗
Stability of process mean and variance; chronological control charts
Statistical basis for checking stability before interpreting capability.
- NIST: Metrology and process control, dealing with measurement uncertainty ↗
Measurement uncertainty and manufacturing risk
Basis for considering uncertainty in acceptance decisions.
Key claims and evidence
| Claim | Basis | Reference |
|---|---|---|
| Quality-system context supports consistent controlled production | documentation | ISO 9001:2026 — Quality management systems: requirements and official publication status |
| Chronological process review identifies changes in behavior | documentation | NIST/SEMATECH: Assessing Process Stability |
| The batch average in the example is 4.06 mΩ | calculation | calculation |
| The record schema is a proposed contractual engineering deliverable | workflow | engineering workflow |