Can we reproduce it?
Trace the claim from source row to formula. Challenge timing, annualization, implementation cost, overlap, and the baseline.
THE RESTAURANT MARGIN EVIDENCE STANDARD
Opportunity is not
savings.
Restaurant groups lose decision quality when signals, models, approvals, implementation, and realized outcomes are compressed into one number. The evidence standard keeps each state separate so Finance and Operations can challenge the claim before capital, labor, or operating risk is committed.
Control boundary: a variance, diagnostic result, modeled opportunity, or approved recommendation is not realized savings. Savings exist only after an implemented control is measured against an approved baseline for an agreed observation window.
Five states. No silent promotion.
Every margin claim should carry one—and only one—current state. Advancement requires the evidence shown below.
- 01Signal observedInvestigate
A variance or operating exception exists. The source, period, location, and affected control are identified. No economic outcome is claimed.
- 02Credible opportunity modeledChallenge
Source records reconcile; formulas, exclusions, confidence, low/base/high cases, overlap, and implementation cost are visible.
- 03Action approvedAuthorize
Finance validates the math. Operations validates feasibility. A named owner, due date, risk, and rollback condition are recorded.
- 04Control implementedObserve
The approved change is live in the bounded cohort. Start date, scope, exceptions, and operational side effects are documented.
- 05Savings realizedReport
The observation window is complete and the result survives baseline, mix, timing, volume, seasonality, and double-count controls.
What each state may claim
| State | Required evidence | Permitted claim | Prohibited shortcut |
|---|---|---|---|
| Signal | Named source, period, location, exception | “A variance requires investigation.” | Calling the variance recoverable cash |
| Modeled | Reconciled sources, formula, confidence, costs, overlap | “The evidence supports a bounded opportunity range.” | Calling annualized opportunity realized savings |
| Approved | Finance and Operations approval, owner, due date, risk | “Management approved a controlled test.” | Calling approval implementation |
| Implemented | Live control, cohort, start date, exceptions, rollback | “The control is operating in the approved cohort.” | Calling early movement a validated result |
| Realized | Completed window, approved baseline, confounder controls, sign-off | “The measured result is realized for this cohort and period.” | Extrapolating beyond the observed scope without a new gate |
Three executive tests
Can we execute it?
Reject actions that damage throughput, food safety, guest experience, labor capacity, training, or local control.
What decision clears?
Stop, extend validation, approve a bounded action, or scale only the result that survived the prior evidence state.
THE MANAGEMENT RULE
Never let a strong model outrank a weak source, an unowned action, or an unfinished observation window.
AI may summarize evidence and draft a narrative. It does not approve the formula, change the operation, or declare savings.
Apply the standard
Use the private diagnostic to test whether the evidence, ownership, execution, and validation controls are strong enough for a bounded pilot. No operating data is uploaded.