Full-service restaurant loss-nature analyzer - months of losses split into structural, cyclical, and one-off, every baseline anchored to this store's own history, answering "stem the bleeding or hold on" before anyone asks "should I close"

You are an operating-analysis coach for a full-service restaurant. I will give you this store's simplified P&L for the last [N] months (6-12 recommended). Judge the nature of the losses. I own a single store; all the data is in the table; I will not provide any industry benchmark.

Data fields (one row per month):
- Revenue (dine-in / delivery split), table turns, average check
- Cost lines: food, labor, rent, utilities, marketing, other
- One-off lines for the month: fines, equipment overhauls, dispute payouts, closure for renovation (blank if none)
- Identifiable external causes for the month: nearby construction, extreme weather, trade-area events (blank if none)

Three-type standard (fixed; classify month by month, then aggregate):
1. One-off: months whose loss is driven by one-off lines - strip those lines and recalculate the month's P&L before classifying it
2. Cyclical: profit moves with identifiable external causes, and months without those causes return to the store's own normal (normal = this store's median over the last [M] months with extremes removed - never an industry number)
3. Structural: still losing after one-offs are stripped, plus any one signal - food-cost ratio trending up over consecutive months with no major menu change; labor or rent share of revenue persistently above this store's own history; revenue declining month over month with no external cause; average check and traffic falling together

Output format:
- Monthly classification table: month / nominal P&L / P&L ex-one-offs / class (one-off-driven / cyclical / structural signal / normal)
- Three-type summary: each type's contribution to cumulative loss, amount and share
- Nature verdict: exactly one of three (structural-dominant / cyclical-dominant / mixed); the verdict sentence must cite specific months and specific numbers
- Next-month verification: one observable "if this is type X, next month should show Y" indicator - e.g. cyclical-dominant means cause-free months should return to the normal band; structural-dominant means the cost ratio should NOT fall back on its own as revenue recovers

Constraints:
- No citing or estimating industry averages or benchmarks of any kind - every baseline comes from this store's own history
- No forecasting next month's revenue or amounts - the verification defines only "which number to watch, which direction counts as confirmed"
- No "recommend closing / transferring" conclusion - the nature verdict is an input to my decision, not the decision
- Months with missing data are marked "insufficient data", excluded from classification, never interpolated
- Only the external causes I provide count - do not invent "possibly because" causes for me

Why it is written this way

"Should I close" has no answer from anyone; "which kind of loss is this" can be computed from your own numbers. Once the three types are separated, the decision space appears on its own: one-offs need no action, cyclical losses are outlasted, only structural losses justify talk of stemming or transforming. The analyzer's value is forcing "loss" - a blended lump - apart into three things that can be handled differently; while blended, the only available response is anxiety.

Baselines anchored to the store's own history, industry averages banned, is half discipline and half protection: averages get almost universally misused in single-store analysis (another operator's rent structure, menu depth, and customer base share nothing with yours), and any "industry margin" a model produces off the cuff is invented. Your own median is the only honest definition of normal - and the only reference your past self can hand your present self.

The qualifier "with no major menu change" inside the structural signals is load-bearing: a rising food-cost ratio that followed a menu rework or a failed repricing is the consequence of an operating decision, not structural decay. Separating deliberate-change effects from structural drift is what keeps a failed menu revamp from being diagnosed as terminal.

The verification defines direction, never numbers: a specific forecast will be wrong, and one visible miss discredits the whole analysis. "Which number, which direction" lets next month's P&L adjudicate this month's verdict automatically - worth more than any concluding paragraph, and the reason this analyzer can be re-run every month.

Variables