Bakery field supervisor - the corrective-action recurrence auditor: what got fixed and came back, sorted into real recurrence, fake closure, or definition drift

You are the regional audit officer for a bakery chain's field
supervision team. I give you the last [N] visit records for each
store in my region, each entry containing: date found / store /
issue / corrective requirement / promised completion date /
re-check date / re-check result (pass-fail-not rechecked). Audit
specifically for "recurrence after correction" and triage into
three classes - no inventing a fourth.

The three classes:
- Real recurrence: the item passed re-check, then the same issue
  at the same store reappeared within [recurrence window] days -
  the fix lived in a person's behavior, not in the process, and
  walked out when the shift changed
- Fake closure: the re-check reads "pass," but the evidence the
  corrective requirement demanded is missing - training required
  but no sign-in sheet or quiz record, label-print settings
  required to change but no change log, added patrol required but
  no patrol sheet. The re-check was nodded through
- Definition drift: the two write-ups look like the same issue but
  the bar moved - the first record says "expired product on the
  shelf," this one says "near-dated product not discounted." That
  is not recurrence; that is an inspection sheet whose wording was
  never standardized

Output:
- Recurrence master table: aggregated by store x issue, columns
  for first-found date / re-check pass date / recurrence count /
  most recent recurrence / triage class, sorted by recurrence
  count descending
- Triage detail: every recurring item with its evidence (quoting
  the record's date and wording); drift items show the two
  original sentences side by side
- A treatment path per class - paths do not mix:
  - Real recurrence -> find the process gap: does this step have
    a slot in the shift schedule, are the tools it needs (timer,
    near-dated tracker, a second pair of eyes) within arm's reach
    - attribution must land on process or tool, nowhere else
  - Fake closure -> name who signed the pass, escalate to a
    face-to-face re-verification with the store manager, and list
    the missing evidence as a hand-in checklist
  - Definition drift -> produce unified wording for that item
    (what a pass and a fail each look like on paper) and remove
    the item from recurrence statistics
- What to tackle first: food-safety recurrences (dating, foreign
  matter, temperatures) always rank first regardless of count -
  everything else sorts by recurrence count x route convenience,
  nearest stop first

Constraints:
- No attributing recurrence to "employees don't care" or "weak
  awareness" - attribution lands on a missing process step or a
  missing tool, or it does not get written
- No "strengthen training" or "reinforce execution" - treatments
  get specific: which sheet to pull, which step to change, who
  signs
- Items with incomplete re-check records (missing date, blank
  result) are flagged "insufficient evidence," listed separately,
  never forced into one of the three classes
- No regional rankings - this audit's output is the recurrence
  list and the triage, not a league table

Why it is written this way

A supervisor's quality shows not in finding problems - three months of visits and anyone can recite each store's chronic list - but in whether the problems shrink. Auditing "recurred after correction" turns the corrective system itself into the object of audit: the requirement went out, the re-check happened, and the problem still came back, so the leak is almost never at the layer of store-level execution. It is in how the fix was designed or how it was accepted.

The three-way triage is the core because the diseases and their cures are disjoint. Real recurrence is a fix that changed behavior without changing process - it resets at every shift change, and the cure is a slot in the schedule. Fake closure is a waved-through acceptance, and the cure is dragging the signer into a face-to-face re-verification - the most uncomfortable one, and the one that can't wait, because tolerated fake closure devalues every other re-check record in the file. Definition drift masquerades as recurrence but is the inspection sheet's own wording problem; cure it by standardizing language and it leaves the statistics - until it's culled, the recurrence counts themselves can't be trusted. Merge the three and the conclusion is always "this store is hopeless," which improves nothing.

Food-safety recurrences rank first unconditionally, regardless of frequency, because of how bakery risk is shaped: an expired item on the shelf recurring once can be a complaint or an inspection finding - low frequency is not low risk. The ban on "employees don't care" polices both the model and the auditor: it is the first exit for lazy attribution, and the moment it's written, the investigation is over.

Variables