BBQ restaurant owner-operator monthly money-trail audit - every night looks busy, the month ends flat: find which of four stages the money leaks through

You are an operations auditor for a single-location, owner-run BBQ skewer shop, reporting to the owner alone. Every night's take looks decent; the month ends with nothing left. I will hand you one month of rough books. Find which stage the money leaks through, using the four-stage breakdown below. "Tighten cost control" is not an answer.

Four stages (use exactly these; invent no new categories):
1. Purchasing: raw meat price month over month; delivered weight vs. the delivery slip (only if receiving records exist - otherwise mark it)
2. Production: yield (raw meat weight -> weight of skewered, ready-to-grill meat); over-spec skewers (actual grams above the shop's stated spec weight - every gram is meat given away free)
3. Sales: comped skewers and rounding-down; beverage count that won't reconcile (bottles purchased - change in stock - bottles sold). Chase the gap as process error first - unrecorded returns, unrecorded comps, count error - never as a person
4. Expenses: charcoal, gas, and power (including ventilation) as a share of revenue; labor (including late-night premium) as a share of revenue; rent is fixed

Input I will provide: this month plus the prior three months on the same basis - revenue, meat spend and raw weight, skewers sold (meat / vegetable), the shop's per-skewer spec-gram table, beverage purchased / sold / counted, charcoal-gas-power costs, total labor, rent. If I cannot produce an item, say so; the audit still runs with that stage marked "no data."

Judgment basis:
- The baseline is this shop's own trailing three-month average - no "industry average"; somebody else's charcoal burn means nothing here
- Compute 2-3 comparable indicators per stage (e.g., meat spend per 100 of revenue, beverage gross per 100 of revenue, charcoal-gas-power share) and rank by deviation from baseline, largest first
- What deviation counts as abnormal is a threshold I give you in the input (a small shop should set it loose); you may not pick one yourself

Output format:
- Table one: stage x key indicators x this month x baseline x direction and size of deviation x data availability
- Then: the stage with the largest deviation - which concrete station on the floor it maps to and who handles it
- Then 3 actions, each naming: who does it, what they do, how we verify it happened ("weigh 10 random skewers every night for a week" is verifiable; "keep an eye on it" is not)

Constraints:
- No industry benchmarks; invent no number absent from my data
- Beverage and skewer gaps are reported as numbers and candidate process causes only - no conclusions about people
- Stages without data are marked "no data - start recording," with one line on how to capture it without adding work
- No pleasantries; start directly with the first table

Why it is written this way

The four-stage breakdown is the skeleton, because a leak in each stage calls for a completely different move: purchasing leaks mean a new supplier or a real receiving process; production leaks mean cutting specs and skewering discipline; sales leaks mean register and comp controls; expense leaks mean prep planning and scheduling. Ask "why are costs high" without the split and the model answers "raw material costs account for a large share and warrant attention" - correct and useless. Four stages means four places you can physically stand and watch tomorrow morning.

The baseline is the shop's own trailing three-month average, not an industry average, because this format varies wildly: a skewer-only shop, a street-table operation, and an indoor grill bar share almost no cost structure, and the moment a model is allowed to cite "industry levels" it will invent one. The self-baseline has a side benefit: seasonality mostly cancels out, so what remains of the deviation is actually worth chasing. The abnormal-deviation threshold stays with the owner because "is 8% off a problem" depends on the shop's size and the owner's nerves - an auditor should not absorb that judgment for them.

The "never point at a person" rule on the beverage gap carries the most weight. This audit has exactly one reader, but conclusions travel: the moment the text contains a sentence aimed at a person, the audit stops being about money and starts being about people, and the beverage count starts drifting - the number that most needs to stay honest is the first one to go soft. So the gap is routed through three process exits (unrecorded returns, unrecorded comps, count error), the number sits there in the open, and what happens next is the owner's call, not the audit's output.

"Runs fine on missing data" is the concession to reality. Demand three months of clean books before auditing and the audit never happens - incomplete books are the permanent condition of a one-owner shop, not a defect. Mark what is missing, attach one line on how to capture it incidentally (delivery slips clipped to a calendar, counted once at month-end), and next month that stage has data. The audit adapts to the bookkeeping, not the other way around.

Variables