This is a ready-to-use form. Complete one sheet per metric and route it for QA approval before the metric is first reported. The definition sheet is the reason two people compute the same metric the same way, and it is the record an inspector asks for when they want to know exactly what a number on your slide means. Replace every <<FILL: ...>> placeholder. A filled specimen for Right First Time follows the blank form. This content is general educational reference, not legal or regulatory advice.
Why this form exists
The hardest part of a quality metrics program is not the chart, it is the definition. Two sites both reporting “OOS rate of 1.2%” can mean completely different things if one counts confirmed OOS and the other counts every initial OOS including lab error. This sheet settles those battles in writing, once, and puts a name against the number. It is the controlled parent record referenced on every dashboard tile.
Metric definition sheet (blank)
| Field | Entry | Notes |
|---|---|---|
| Metric name | <<FILL: e.g. Right First Time (batch records)>> | The exact display name used on the dashboard |
| Metric ID | <<FILL: MDS-###>> | Unique, version-controlled |
| Version / effective date | <<FILL: v1.0 / date>> | Under change control |
| Owner (role) | <<FILL: QA SME role>> | Accountable for calculation and narrative |
| Classification | Leading / Lagging / Mixed | Leading predicts, lagging confirms |
| Plain-language definition | <<FILL: one sentence a non-specialist understands>> | |
| Formula | <<FILL: numerator / denominator x 100, stated exactly>> | |
| Numerator rule | <<FILL: what counts, what one unit is>> | |
| Denominator rule | <<FILL: the normalizing base that scales with the numerator>> | e.g. per 100 batches, per million units |
| Inclusions | <<FILL: what is counted>> | |
| Exclusions (with written rule) | <<FILL: what is not counted, and the pre-defined rule>> | No undocumented exclusions |
| Multiple-defect handling | <<FILL: any defect fails the unit / count each defect>> | Standard: any defect fails the unit |
| Source system of record | <<FILL: QMS / LIMS / LMS / disposition log>> | Single validated source |
| Data pull method | Programmatic / validated spreadsheet / manual | Manual pulls require reconciliation |
| Reporting frequency | <<FILL: monthly / quarterly>> | |
| Target (goal) + rationale | <<FILL: value + why (baseline, benchmark)>> | Never below current performance |
| Action threshold (limit) | <<FILL: value that forces a documented response>> | |
| Trend rule | <<FILL: e.g. 7 points one side of mean; 3 months decline>> | Independent of any single point |
| Paired metric(s) | <<FILL: the metric(s) read alongside this one>> | e.g. RFT paired with deviation rate |
| Related SOP / WI | <<FILL: the procedure that governs the process measured>> | |
| Approved by (QA) | <<FILL: name, signature, date>> |
Field instructions
- Classification. Decide leading, lagging, or mixed before anything else. A program that leads with lagging metrics is watching the crash in the rear-view mirror.
- Denominator rule. Pick the base that scales with the same driver as the numerator. If production halves and a raw monthly count halves with no real quality change, the metric lies; a per-batch or per-unit rate holds steady.
- Exclusions. Every exclusion needs a written, pre-defined rule. “We dropped that month, it was an outlier” with no rule reads as manipulation.
- Target rationale. “95% because the trailing 24-month mean is 94% and we want to drive improvement” is defensible; “95% because it is a nice number” is not, and an inspector will ask.
- Trend rule. Use a run rule, not a reaction to single points, so signal is separated from noise on your most important metrics.
Retention
Retain each version of the definition sheet for the life of the metric plus <<FILL: retention period>>, so an inspector can reconstruct exactly how the metric was defined during any historical period, including across definition changes.
Filled specimen
The following shows the sheet completed for Right First Time on batch records. The numbers and IDs are illustrative; replace them with your own.
| Field | Entry |
|---|---|
| Metric name | Right First Time (batch records) |
| Metric ID | MDS-014 |
| Version / effective date | v2.0 / 01 July 2026 |
| Owner (role) | QA Disposition Lead |
| Classification | Leading |
| Plain-language definition | The percentage of batch records that reach disposition with no documentation defect requiring correction |
| Formula | (Batch records with zero defects / total batch records reviewed at disposition) x 100 |
| Numerator rule | A record with any defect (a correction requiring second-person check, a missing entry, a deviation raised against it, a re-test for analyst error) fails; one unit is one batch record |
| Denominator rule | All batch records reviewed at disposition in the period |
| Inclusions | All GMP batch records dispositioned in the period |
| Exclusions (with written rule) | Records for engineering / non-GMP runs, per rule in SOP-QA-014 s4.2 |
| Multiple-defect handling | Any defect fails the unit; a record with three errors is one RFT miss, not three |
| Source system of record | Electronic batch record system / disposition log |
| Data pull method | Programmatic |
| Reporting frequency | Monthly |
| Target (goal) + rationale | 95%; trailing 24-month mean is 94%, target set to drive improvement |
| Action threshold (limit) | Below 90% forces a management action |
| Trend rule | Three consecutive months of decline, or seven points on one side of the mean |
| Paired metric(s) | Deviation rate; invalidated OOS fraction |
| Related SOP / WI | WI-QA-014 (batch record review) |
| Approved by (QA) | R. Gomez, signed, 01 July 2026 |
In a month where 220 batch records are reviewed and 31 have at least one defect, this sheet resolves the calculation unambiguously: RFT = (220 - 31) / 220 x 100 = 85.9%, one miss per defective record regardless of how many defects it carried, engineering runs excluded. Anyone recomputing from the same source gets the same number.
Common inspection findings this form prevents
- The same metric computed two different ways by two people, because the counting rule was never written down.
- A trend that steps because someone quietly changed what counts, with no version history to show it.
- Targets no one can justify when challenged.
- Undocumented “outlier” exclusions that look like a metric being gamed.
- A dashboard tile with no owner, no source system, and no traceable definition behind it.
How to adapt this form
- Create one sheet per metric in your mandatory set and give each a unique ID.
- Put the source system and the plain-language definition on the corresponding dashboard tile so the number is inspection-ready at a glance.
- Route each sheet through change control; changing a definition mid-year is allowed but must be annotated on the trend.
- Keep superseded versions so historical trends remain interpretable.