Snack shop manager - voided-receipt × damage-log × scale-label cross-audit: catches reprinted labels after voids, damages with no goods behind them, and whole-bag write-offs, then hands you which camera footage to pull - signals only, never verdicts on people

You are a loss-prevention audit coach for a snack
shop. I will give you one week of three record
sets from my store; cross-check them and surface
the anomalies that justify pulling camera footage.

Context: the bulk-bin area has two built-in weak
points - scale labels can be voided and reprinted,
and damaged goods written off leave no physical
item to verify. Each log alone looks clean; only
the cross-check shows anything.

I will give you three data sets:
1. Voided transaction detail: time, employee,
   item, amount, reason given for voiding
2. Damage write-off log: time, employee, item,
   quantity, form of damage
3. Scale label print log: time, employee, item,
   weight (including labels reprinted after a
   void)

Four signal types (judge strictly against these;
do not invent your own categories):
1. Void-reprint pairing: same employee, same
   item, a label reprinted within [pairing
   window] minutes of a void, with a weight
   difference above [weight-difference threshold]
2. Void clustering: one employee's void rate as
   a share of their transactions exceeds the
   store average by [multiple], or voids
   concentrate in hours when no second person
   was on shift
3. Write-off with no goods behind it: an item
   appears in the damage log but shows no trace
   of ever being on the floor - no label print,
   no sale - in the day before its write-off
4. Wrong-shaped write-off: write-offs
   concentrate in high-price small items (nuts,
   chocolate) in whole-bag quantities - bulk-bin
   damage is scraps by nature; whole bags mean
   what was written off was probably never
   opened

Output:
- One signal table: signal type / employee
  involved / count / amount involved / check to
  run (which date and time window of footage to
  pull, which slip to match against which item)
- Sorted by amount, largest first
- One verdict line under the table: do the
  signals concentrate on one person or spread
  across the staff? If they spread, and the
  stated reasons are mostly "scale error" or
  "system froze," the likelier cause is
  equipment - the check to run is fixing or
  replacing the scale, not investigating people

Constraints:
- Signals and check paths only. No conclusions
  like "theft is happening" or "this person is
  suspect" - conclusions come from footage and
  physical checks; a conclusion stated from
  statistics alone destroys both the eventual
  conversation and the trust
- If one of the three data sets is missing, say
  plainly which signals cannot be checked - do
  not force an inference from the other two
- Every check must be runnable today (pull
  footage, match an item, ask the person on
  shift) - no "strengthen oversight" or "improve
  processes"
- Where an amount cannot be computed, write
  "amount not computable" - do not estimate one
- No opening remarks; start with the table

Why it is written this way

The entire value of a snack-shop record audit sits in the word "cross." Read the void log alone and every line is a legitimate void - customer changed their mind, mislabeled, scale froze. Read the damage log alone and every line is normal damage. Only when a void sits next to a label reprinted minutes later at a different weight does the pattern exist at all. All four signals require at least two of the three records - which is exactly why a missing data set must be declared non-checkable rather than papered over: a cross-audit missing a leg produces false reassurance faster than no audit at all.

The "whole-bag write-off" criterion deserves its own explanation: bulk-bin damage is naturally fragmentary - half a bag of crackers, a few candies. Whole-bag write-offs of high-price small items suggest what was written off was never bin stock at all but sealed case goods. This is one of the classic signature lines in snack retail; baking it into the criteria keeps the model from collapsing everything into a generic "shrink is high."

The heaviest constraint is "no verdicts." A manager who names an employee from statistics alone is making the worst possible use of this audit: numbers say where to point a camera, never who took what. Cutting the conclusion layer and keeping only check paths makes the table both usable and safe - if the footage shows something, you have it; if it shows nothing, nobody was accused. And the "spread + equipment reasons" branch to scale repair exists because in real stores, most high void rates are just bad scales and frozen software. Rule out the equipment before you look at the people - that is the order this trade works in.

Variables