Store off-system purchasing early warning for field supervisors - order gap, category zero-out, and material-origin signals flag only the stores whose numbers do not add up

You are a regional field supervisor in a central kitchen
system. I will give you order and sales data for the stores
in my territory over the last [N] weeks. Identify the stores
whose order gap cannot be explained - the ones that may be
buying outside the distribution system. This is a warning,
not a verdict.

Input per store:
- Weekly order quantities from the central kitchen, item
  by item for the core 20 items
- The same weeks' sales (store revenue or main-category
  sales)
- Return rate, store-reported shrinkage rate
- Events I annotate: manager change, renovation, a
  competitor opening nearby

Three signals (more hits, higher the level):
1. Gap signal: sales flat or rising while core-item orders
   fall for [3] consecutive weeks or more - demand
   unchanged, supply channel changed
2. Structure signal: total order volume normal but
   specific items at zero - especially items whose raw
   materials are easy to source locally and carry high
   store margin (prepared dishes and fresh-cut vegetables
   first)
3. Material signal: my walk-through notes show non-system
  packaging; menu items that are not in the distribution
   catalog; return rate abnormally high (unstable off-
   system quality usually shows up as returns)

Warning levels:
- Red: all three signals hit, or the material signal hits
  alone (physical evidence stands on its own)
- Yellow: two signals hit
- Watch: gap signal only, and explainable by an event I
  annotated - watch-level stores do not enter the report
  body, only the appendix

Next-step verification for red stores (written for the
supervisor, not to be sent to the store):
- What to check: physical inventory of non-system items
  in the walk-in, with photos; the store's receiving
  logs reconciled day by day against central-kitchen
  outbound records; the monthly gap between POS item
  sales and order volume
- How to check without triggering a confrontation: enter
  under a routine-inventory pretext, secure the evidence
  first, then schedule the talk. Do not walk in asking
  "are you buying your own produce"

Output:
- Table 1: store by store, signals hit and warning level
  (red / yellow / watch); red and yellow stores each get
  one row of evidence chain
- Table 2: the verification path for red stores - the
  three checks above expanded into executable steps with
  deadlines
- Appendix: watch-level stores and their explaining
  events, one line each

Constraints:
- Output only "gaps that cannot be explained"; do not
  conclude "this store is buying off-system" - the
  warning's job is who to check and what to check; the
  verdict comes from verification evidence
- For data I did not supply (POS line detail), state that
  it is excluded from this judgment; do not fill gaps
  from imagination
- Do not recommend "re-educating stores on distribution
  policy" - this is a verification problem, not an
  attitude problem

Why it is written this way

Warning without verdict is the first design decision. The classic supervisor mistake is marching into a store armed with a falling order curve, only to find the store ran a promotion or shifted its menu mix - trust is wrecked in one visit, and the stores actually buying off-system learn to square their books. All three signals are evidence of "worth checking," never evidence of guilt; even red stores receive a verification path, not an accusation. Complete the three checks - walk-in inventory, day-by-day document reconciliation, sales-gap math - and the verdict establishes itself without the warning overreaching.

Sales as the control variable is what keeps the gap signal from firing on innocent stores. Order volume alone indicts every store with genuinely declining traffic; the standard is "sales flat or rising while orders fall" - only when the demand side is stable does a changed supply side become suspicious. The structure signal watches for category zero-out rather than total decline because experienced side-buyers never replace everything; they cherry-pick the items with easy local supply and the best margin, and the total barely moves.

The explanation channel gives the data an appeals process. A manager change, a renovation, a new competitor can all legitimately produce a gap - annotated events downgrade those stores to watch level and out of the report body. This is how the report keeps its false-positive rate survivable: flood readers with wrong flags twice, and by the third round nobody opens the report. Off-system detection is a long game; credibility is worth more than sensitivity.

The verification path's "enter on a routine-inventory pretext" comes straight from field lessons: a supervisor walking in with an accusation finds the walk-in cleaned out that same night and the evidence gone. Photos and documents first, confrontation second - that sequence is about holding the evidence, not about tricking anyone. Every step names its deliverable (photos, day-by-day reconciliation, monthly gap), so when the talk finally happens, the supervisor is holding facts.

Variables