Verification and standards
We are designing NutriChoice around restaurant-supplied recipe records, documented checks and traceable updates. This page explains the process we intend to test before public ordering begins.
Planned process. Pilot verification has not yet been completed.
Begin with the recipe
Ingredient records, defined portions and preparation notes are the basis of the planned process.
Keep versions connected
A change in recipe, supplier or portion should prompt a review of affected information.
Make uncertainty visible
Show the input and the reason. Withhold a comparison when relevant information is missing.
What the labels on this site mean
Done: something that has happened, shown with its date. Designed: what our plan requires; it is not operating yet. Planned or targeted: a target date and what it depends on. Sample: fictional, for illustration only.
How often records are to be checked
| Check | When |
|---|---|
| Remote check against a fixed list of evidence | Every quarter. |
| On-site audit | Within 90 days of going live, then at least once a year. |
| Recipe change | Before the changed dish can be offered. |
By design, a location cannot take its first order until 100% of its menu is labelled, and a menu change goes back through the same steps before it appears.
When a record is out of date or incomplete
A recipe record that is not confirmed again on time is designed to go out of date automatically, and its dish leaves the checked views until the restaurant confirms it.
When a figure is missing, the dish is held back from the views that need that figure and is never shown as a match.
Four written procedures
Four written procedures are planned to govern the service from the first pilot order: privacy; allergens and labelling; verification and quality checks; and incident response.
Start with the restaurant’s own records
Each proposed dish record includes ingredients, quantities, sub-ingredients, portion size and preparation notes. Sauces, toppings and substitutions need their own information because they can change what is in the meal.
Make each check understandable
Restaurant confirmation establishes who supplied a record. A document review checks the information against the relevant requirements. An on-site review examines applicable preparation practices. These are different checks; the published record should say which ones have actually been completed.
Keep screening separate from recommendations
The planned screening rules compare recorded meal information with selected dietary limits. Preference ranking then helps organise the eligible results. Ranking must not override a screening rule or turn missing information into a positive result.
Show uncertainty
If an ingredient quantity or preparation detail is missing, the affected result should be marked unavailable or withheld from the relevant screened view. We do not intend to infer suitability from a dish name or photograph.
Review changes, not just new listings
The proposed process records menu versions and calls for renewed review when ingredients, suppliers, portions or preparation practices change. Completed reviews and outstanding issues should remain distinguishable.
What a listing cannot promise
Ingredient information alone cannot establish that a meal is free from cross-contact. Preparation information has limits, and a screening result is not a guarantee. NutriChoice is being developed to inform food choices; it does not diagnose, prescribe or replace advice from your healthcare professional.
We will update this page as the process is tested and the scope of completed checks becomes clear.
What we plan to report from the pilot
Each pilot location’s monthly report is planned to cover orders, service levels, refunds, labelling completeness and customer satisfaction, signed off by the location and by the restaurant group’s head office. We plan to publish these measures, with the number of orders behind each one, once they exist.
A record should be inspectable
Our planned information model separates the source record, the checks performed and the explanation shown to a diner. The goal is to make it possible to understand a comparison without treating a platform label as a blanket assurance.
A review history should identify the recipe version and what changed. When relevant data is unavailable, the affected comparison should be withheld or qualified. The current website demonstrates this behaviour with fictional sample records only.