What the current ChefQuote runtime actually does.
ChefQuote's current daily and operating-memory functions use deterministic decision-support logic. They do not call a generative model or present themselves as a human. This page records the present behavior, safeguards, and the conditions that require a new assessment under Regulation (EU) 2024/1689 and other applicable law. Last reviewed 4 August 2026.
Fixed decision-support automation
ChefQuote uses deterministic operating logic to prepare estimates, planning reviews, schedule-exposure counts, same-service memory checks, and a prioritized daily operating read. It does not convert schedules into morale or health scores. The runtime does not call a generative model, train itself, or silently change its protected logic.
A reasoned human decision is required
No memory recommendation is applied on its own. Accept, adjust, and reject each require a written operator reason, and the action is logged with a protected method-release label, evidence record IDs, before/after values, safeguards, and timestamp.
No biometric input in this build
ChefQuote does not request camera, microphone, voice, facial, keystroke, wearable, or heart-rate signals. Adding a biometric or emotion-inference input is a release red line requiring new design, legal review, and a new compliance assessment; it is not authorized by the current product scope.
Not an employment decision tool
ChefQuote does not screen, rank, or score job candidates, and produces no discipline, demotion, or termination recommendation.
Owner-visible control
The owner can inspect the business-facing reasons, labels, action explanations, review settings, and saved change records. Proprietary calculation and ranking mechanics remain protected. Saving or resetting owner-facing configuration requires a reason. That control does not certify an owner change as safe or lawful.
What “closed-loop operating system” means
The phrase describes ChefQuote's connected operating workflow: decide, execute, measure, review, and return an operator-approved lesson. It does not mean a device operating system, a self-modifying AI model, autonomous management, or guaranteed improvement.
Why crew signals are not emotion recognition.
The AI Act prohibits certain AI systems used to infer a worker's emotions in the workplace, subject to its stated medical or safety exception. The current product does not use biometric data or claim to detect emotion. Classification still turns on actual intended purpose, inputs, outputs, and use — not this page's label.
ChefQuote currently uses two non-biometric source classes. Class A is operator-entered or schedule-derived operational data such as shift length, consecutive days, turnaround, overtime, and workload concentration. Class B enters the production intake only as an operator-provided team-level survey aggregate. Outputs keep operational exposure separate from direct aggregate self-report; they do not claim to detect emotional state. This design reduces Article 5(1)(f) risk, but it is not a legal determination that every deployment is outside the prohibition.
Crew features keep stricter boundaries.
Annex III addresses certain AI systems intended for employment and workers' management. ChefQuote's current deterministic runtime is not classified as a high-risk AI system by this page. Because workforce-related uses can still affect people, the product applies conservative controls and requires a fresh legal and technical assessment before any learned model, individual worker evaluation, employment recommendation, biometric input, or materially changed intended purpose is released.
Risk management & data governance
Schedule imports produce operating counts only. Manager planning categories are framed as unvalidated product judgment, and direct survey dimensions are never collapsed into a morale, health, or retention score.
Human oversight by design
The structured intake rejects individual survey rows and cohorts below five. The separate questionnaire is a static session-only prototype, not a production aggregation service.
Records & logging
Changes to quoting logic, people data handling, and public claims are recorded with the date and the provision they touch.
Transparency to the affected person
Crew are told what is collected, what it feeds, and the list of things it will never be used for, before they answer anything.
AI-assisted coding is not the product classification test.
Using an AI assistant to write or edit source code does not by itself make the deployed website an AI system. Conversely, replacing AI-written code with human-written code does not create an exemption if the runtime functionality still falls within the Act's definition or a regulated intended use. ChefQuote's present daily and memory engines are described publicly as deterministic decision support, while implementation details remain protected. Any future model, learned parameter, biometric input, individual worker evaluation, or materially changed intended purpose triggers a new assessment before release.
What this page is not.
This is a good-faith transparency notice, not legal advice and not a certification. A formal conformity assessment, EU database registration, and any authorised-representative requirement are matters for qualified counsel. If you deploy ChefQuote with staff in the EU, GDPR and national labour and works-council law may apply alongside the AI Act. Controller and processor roles depend on the actual deployment and contracts; this build does not determine them. Have counsel review before staff use.
Primary references checked 4 August 2026: Regulation (EU) 2024/1689 on EUR-Lex and the European Commission's Article 50 transparency guidelines. Article 50 notice must be clear and distinguishable at first interaction or exposure when its covered conditions apply. This testing build does not treat a footer page as a substitute for point-of-use notice if a future qualifying AI interaction or covered synthetic-content feature is introduced.
Import automation is constrained data processing, not an autonomous worker-management system.
ChefQuote maps and validates CSV fields with fixed rules. It does not use a general-purpose model to infer missing labor, cost, schedule, or survey values. Unsupported PDFs are held instead of guessed.
No individual survey import
Names, emails, worker IDs, comments, free text, and employee cohorts below five are rejected. Only approved aggregate results can enter the crew read.
No individual employment output
Schedule rows are reduced locally to an approved team aggregate; raw shifts are not retained by this path. The output cannot be used for hiring, firing, discipline, task allocation against a person, performance scoring, or reducing hours.
Human review is a real gate
Required checks and a source control total must pass. Accept, adjust, and reject require a written reason. A source hash and change record remain available for audit.
This design reduces risk; it is not a legal determination that every deployment or later modification falls outside the EU AI Act, employment law, privacy law, works-council duties, or state and local worker-monitoring rules. Intended use and actual use still control the legal analysis.