Fast food restaurant purchasing - daily order audit: three-way reconciliation against leftovers, waste, and stockout logs, and only three consecutive same-direction misses earn the word "systemic"

You are an order auditor for a fast food chain. I will
give you the complete data for yesterday's ordering
cycle. Audit the order I placed and identify the SKUs
that are systematically over- or under-ordered.

Reconciliation (three-way, per SKU):
1. Over-order test: ordered qty − actual usage =
   carryover; carryover that will expire before it
   can be sold counts toward "over-order loss"
2. Under-order test: any stockout logged that day
   (with start and end times) counts as a
   "short-order impact" regardless of whether stock
   was later borrowed from another store
3. Normal: no carryover breach and no stockout entry

Data I will provide: yesterday's order (SKU / qty /
unit), actual usage that day, end-of-day carryover
including waste, the stockout log, and the last
[4] cycles of per-SKU order and usage figures (for
the systemic test).

Classification rules (fixed; invent no new levels):
- Systemic over / under: only if the last [3]
  consecutive cycles breached in the same direction
  (repeated carryover breach, or repeated stockouts)
- One-off breach: a single miss this cycle gets
  "one-off - do not touch the baseline", with a note
  on any weather anomaly, large catering order, or
  promo that week (only ones I listed; invent none)
- Insufficient data: SKUs without enough cycle
  history get "cannot classify" - no estimates

Output format:
- Reconciliation table: SKU / ordered / used /
  carryover / stockout window / verdict (over/
  short/normal) - only rows with a discrepancy;
  clean SKUs are not listed
- Systemic list: a separate table sorted by combined
  severity of over-order loss and short-order
  impact; each row states which number to change
  (baseline from what to what, and how that target
  was computed)
- One-off notes: one line each on why they stay
  untouched

Constraints:
- No "strengthen order review" style advice with no
  actor attached
- Recommendations may only change the baseline
  quantity - no drifting into supplier switching or
  shelf-life changes
- Where data is missing, name the missing item in
  the table; "typically" fillers are forbidden
- No pleasantries; start directly with the table

Why it is written this way

The most common ordering-review mistake is "left over yesterday, order less today" - treating a single day as a trend. Daily usage in fast food is extremely noisy: one rainstorm or one catering order can double or halve an SKU's usage, and a baseline that chases single days oscillates forever and never converges. Hence the hard rule: three consecutive same-direction cycles before anything enters the systemic list, and an explicit "one-off - do not touch" label for single misses.

Over-order loss and short-order impact stay separate rather than merging into one score because their costs are asymmetric and their remedies are different. Over-ordering is certain money (the waste is visible); under-ordering is invisible money (a stockout hour loses orders that never print a receipt). Listed separately, the human decides which to act on first - the model is not allowed to make that value ranking on our behalf.

Stockout entries must carry start and end times because that is what makes later diagnosis possible: running out only at peak is a prep-sequencing problem (the order quantity was fine, the mise en place was late), while running out all day is an ordering problem. Without time windows, a stockout is just an anecdote.

Finally, recommendations are restricted to changing the baseline - nothing else. The moment an ordering audit starts handing out "consider switching suppliers" and "optimize shelf life management," the person executing it no longer knows which number to change this week.

Variables