Record defects without changing the denominator

Record defects without changing the denominator

A useful check sheet makes the unit of observation visible. One rejected product can contain several distinct nonconformities, so a tally of faults is not automatically a count of failed products. Define each category, the inspection opportunity and the observation window before recording. Pilot the sheet with more than one observer and resolve ambiguous classifications. Keep identifiers or source labels that enable a later investigation without collecting unnecessary personal data. Record the denominator alongside the tally. Only then calculate a rate or compare shifts; a larger count may simply reflect a larger amount inspected.

Follow the method

1. Unit record / multiple faults allowed
2. Category tally
3. 6 / 50 units = 12% nonconforming
4. 8 / 50 = 0.16 faults per unit

Reconcile units and fault marks before reporting a rate

Fictional case: an incoming-inspection team checks 50 covers during one defined receipt window. The core illustration retains six nonconforming units and eight nonconformities. Inspector Arun must produce a record that another inspector can reproduce. The supervisor has called both numbers “rejects”, making the reported rate ambiguous.

Follow the method

1. Nonconforming units
2. Nonconformities per unit
3. Category reconciliation

Six affected units, eight recorded faults

Record one inspection consistently and reconcile nonconforming units separately from the total number of nonconformities.

Fictional case: an incoming-inspection team checks 50 covers during one defined receipt window. The core illustration retains six nonconforming units and eight nonconformities. Inspector Arun must produce a record that another inspector can reproduce. The supervisor has called both numbers “rejects”, making the reported rate ambiguous.

Role: Incoming inspector and quality record owner

Normal condition

Each inspected cover has one identity; every observed category is recorded once under an agreed definition; total inspected units and affected-unit count remain explicit.

The gap

U01 and U02 each have two different faults. Counting every mark as a different rejected cover would exaggerate the fraction of units affected.

• This is a complete fictional 50-unit inspection record, not a sampling acceptance plan.
• Category definitions describe observed conditions; they do not prove process causes.

1. Agree on the observable definitions

   Arun defines the inspected unit as one cover and clarifies the scratch, burr, missing-label and other categories with examples. The team records uncertainty for adjudication instead of guessing a category.

   Why: A neat tally is unreliable when inspectors apply different definitions. A check sheet is a collection form; ticking a procedural checklist is a different use.

   Evidence: The sheet states unit, window, categories and the rule for ambiguous observations.

2. Pilot the recording method

   Two inspectors independently classify a small set before live collection. They discuss ambiguous examples and revise wording before the receipt window starts.

   Why: The pilot exposes unusable categories or insufficient writing space. Agreement during discussion does not by itself establish a complete measurement-system study.

   Evidence: The released form has a revision and the agreed observation instructions.

3. Keep unit identity and multiple marks

   Arun enters U01 through U06 once each and retains both categories on U01 and U02. He records the total population inspected, including the 44 with no recorded faults.

   Why: Fault tallies without the inspected population cannot yield a unit rate. Multiple faults on one unit are permitted and must not create duplicate units.

   Evidence: Six distinct affected IDs; eight category marks; fifty inspected IDs accounted for.

4. Reconcile before calculating

   He sums category counts: scratches 3, burrs 2, missing labels 2 and other 1. He separately counts the six unique affected units.

   Why: The sum of category marks answers how many nonconformities were observed. A distinct-unit count answers how many inspected units were affected.

   Evidence: 3 + 2 + 2 + 1 = 8 faults; six unique nonconforming units.

5. Publish two labelled measures

   The report states 6/50 = 12% nonconforming units and 8/50 = 0.16 nonconformities per inspected unit. Both retain the window and inspection scope.

   Why: These rates answer different questions. Neither says that 16% of units failed, and neither establishes a process trend from this one window.

   Evidence: A reviewer can trace each numerator and denominator to the same source sheet.

A scratch is entered twice

A second transcription of U01’s same scratch appears in the summary.

Reconcile against U01’s original observation, correct the duplicate with a traceable amendment, and preserve the legitimate burr entry.

Correcting a duplicate is not deleting an inconvenient result. The source record and amendment explain why the total changes.

Corrected category count refers to one scratch observation on U01, not two.

Run a practice session

Materials

• Blank unit-by-category grid
• Six fictional unit cards
• Separate receipt exercise

1. Define a unit · 5 minutes

   Can one cover contribute two marks?

2. Demonstrate the tally · 7 minutes

   Keep U01 visible while calculating both numerators.

3. Reconcile the changed receipt · 9 minutes

   Where does each of the six marks come from?

4. Debrief · 5 minutes

   What information would a tally-only summary lose?

Debrief

• Ask learners to explain 10% and 0.15 in ordinary language.
• Use an ambiguous scratch to discuss definition quality rather than force agreement.

Draw the four affected practice units before doing the arithmetic.

Transfer into the work

Owner: Inspection process owner

Record: Released collection form and reconciled summary

Review: After the pilot and at the next scheduled data review

Evidence: Different reviewers reproduce the counts from the same original records

Correct definitions, training or record capture before comparing rates across windows.