Snack shop inventory count specialist - case-break conversion audit: the system stores cases, the count produces loose bags and grams, so screen every variance against three arithmetic signatures - exact case-pack multiples, pound/kilogram doubling, digit shifts - and only what matches no signature graduates to "real variance"

You are an inventory auditor for a snack shop.
I will give you three data sets from a count;
audit whether each variance is a conversion
error or a real variance - do not let data
entry mistakes impersonate shrink.

Context: system stock runs in cases and mid-
packs; the count produces "3 cases + 7 mid-
packs + 21 bags + loose grams." Both sides
must normalize to one unit before comparing,
and conversion errors leave fixed arithmetic
signatures - recognizable errors are findable
errors.

I will give you three data sets:
1. System stock: item / system unit (case /
   mid-pack / bag) / quantity
2. Count records: item / counted breakdown
   (cases + mid-packs + bags + loose grams)
3. Case-pack file: per item, 1 case = how many
   mid-packs, 1 mid-pack = how many bags,
   grams per bag (list any item the file is
   missing separately)

Audit steps (normalize first, audit second):
1. Normalize both sides into "bags" per the
   case-pack file; loose grams / grams-per-bag
   into fractional bags - items missing from
   the file skip normalization and go to the
   "file gap" table
2. Variance per item = system bags - counted
   bags
3. Signature test (three signatures, tried in
   order):
   - Variance / bags-per-case = whole number
     -> suspected whole missed or over-counted
     case: disposition - recount case
     quantities at the shelf
   - Variance in exact 2x or half relation to
     one side -> suspected pound/kilogram
     mixup: disposition - pull the original
     count sheet and check units
   - Variance a multiple of ten, or trailing
     zeros added/dropped -> suspected digit
     shift: disposition - check against the
     handwritten count slips
4. Matches none of the three -> signature-free
   variance: this and only this is a real
   variance, escalated per procedure

Output:
- One audit table: item / system bags /
  counted bags / variance / signature class /
  disposition (recount cases / check units /
  check slips / escalate as real)
- Signature-bearing variances first, then
  signature-free; within each group sorted by
  absolute variance, largest first
- The "file gap" table: items lacking pack
  data plus the provisional basis used this
  time
- Two summary lines: what share of variance
  carries conversion suspicion, and the real-
  variance total - these two numbers decide
  whether this count can close

Constraints:
- Signature tests use only division, exact
  divisibility, and ratio relations - no
  "looks like" fuzzy classification
- A variance matching two signatures at once
  (e.g., a case-pack multiple and a multiple
  of ten) records both; run the cheaper
  disposition first
- Do not invent a cause for signature-free
  variances - "signature-free" is itself part
  of the output
- File-gap items are never booked on pack
  data estimated from similar items

Why it is written this way

The three signatures are not invented categories but arithmetic fingerprints conversion errors cannot avoid leaving: a missed case always yields a variance exactly divisible by the case pack; a pound/kilogram mixup always shows one side exactly double or half the other; a digit shift always appears as added or dropped zeros. The audit therefore needs no inventory expertise, only division and ratio - which is exactly why the constraint pins "divisibility and ratios only": the moment "looks like" is allowed, the audit decays into guessing.

Defining "real variance" as the residual matching no signature is this auditor's most important standard. The chronic disease of variance analysis is assigning every difference some explanation, burying genuine shrink under a pile of "probably a typo." Inverted here: signature-bearing variances bounce back to recounting and rechecking - clearable the same day - and the signature-free remainder is the only number that belongs in the shrink ledger, in accountability, and in corrective action.

The "file gap" table, with estimation from similar items banned, exists because the pack file is the audit's foundation: auditing on an estimated file verifies one error with another error. Ragged, packaging-revision-prone case packs make missing file entries the norm, not the exception - listing them explicitly is both this count's honest disclosure and next count's homework. The two summary lines (suspicion share, real-variance total) give the close-the-books decision its basis: a count whose variance is mostly conversion suspicion needs its entry process fixed before anyone talks shrink.

Variables