Technical reference
says what happened. tests whether monitoring can be believed. A sample makes that affordable, as long as the person checking is independent and the sample isn't chosen for comfort.
Record sampling and lot sampling answer different questions. A lot decides whether food is acceptable. Record sampling asks whether the system can be believed. The two use the same word and very different logic.
A simple rule from statistics helps here. If none of n randomly chosen records is faulty, the true fault rate could still be as high as about 3/n at 95% confidence. So 10 clean records only suggests under about 30%, and 30 clean records suggests under about 10%. A sample chosen for risk finds faults faster, but can't be turned into a rate this way.
Parameters
| Who checks | Someone other than the person who and corrected |
|---|---|
| How many | Set in your plan. sets no number |
| Period | Day, week, month or quarter, per process and work area |
| What counts | A completed check of the right type, inside the period, passed or failed |
| Original record | Never changed by a check |
| A failed check | Needs food action, cause and before it closes |
Records that prove it
- Sampling plan, dated, with each change
- Required, done and outstanding, per period
- Each check: record checked, who, when, own reading, result
- Failed checks: food action, cause, corrective action
- Review of non-compliant records, by someone who did not record them
CCP or prerequisite?
That depends on the for your process. The number, if it has one, belongs to your plan.