Research method
We focus on transparency and traceability. The goal is to make each rule easy to audit, discuss, and revise as evidence or guidance evolves.
See also: Responsible use & limitations
How to read this page
- • The system uses rule-based logic that can be reviewed and explained at each step.
- • Outputs are informational and designed to support understanding, not to diagnose or treat.
- • The product is in active beta, with ongoing refinement based on feedback and evidence.
Sources and governance
Rules are grounded in public nutrition labeling guidance and published thresholds. We track sources, version changes, and open questions to keep the logic reviewable.
Rule design
Each rule is explicit, with defined inputs and output language. We avoid probabilistic scoring and keep the logic human-readable for partner review.
Evaluation approach
- Usability testing to confirm clarity of explanations.
- Expert review of rule mapping and thresholds.
- Pilot feedback from partners to understand real-world constraints.
Primary rule prioritisation
Users often have one dominant dietary requirement—such as avoiding gluten, managing sodium intake, or reducing sugar.
The system prioritises this primary rule in evaluations, treating secondary preferences as contextual rather than equally weighted. This reflects how people make food decisions in practice.
This is a design decision intended to reduce cognitive load and improve clarity. It is not a medical judgement about which dietary needs are more important.
Explainability in Practice
Transparency, limitations, and non-diagnostic intent
Screens shown reflect the current beta implementation and may change.
Evidence Summary (Current Beta)
The following statements describe what is implemented and observable in the current beta release.
- Rule-based evaluation logic is implemented.
- Preference weighting is operational.
- Explainability is surfaced in the interface.
- Confidence indicators are shown where data is limited.
- The system is already in active beta use.
The current beta does not claim or attempt the following:
- No medical diagnosis.
- No individualised medical advice.
- No predictive health outcomes.
- No replacement for professional guidance.
Current Limitations
This scope is intentionally constrained to keep outputs reviewable and interpretable during beta.
- Data coverage varies by product and region.
- Ingredient completeness is not guaranteed.
- Conservative warnings may appear in ambiguous cases.
- Evaluation logic is designed to prioritise transparency over certainty.
- Ongoing iteration and validation are expected during beta.
Explainability before model tuning
MyFoodFit prioritises transparency over prediction accuracy. The system uses explainable logic rather than complex models that cannot be easily audited or understood.
This means that sophistication is deliberately limited in favour of interpretability. Users and reviewers can trace each evaluation back to specific rules and data points.
Where data is uncertain or incomplete, the system signals this conservatively rather than inferring or predicting missing information.
This approach reflects the responsibility of operating in a domain where clarity and caution are more appropriate than confidence.
Development principles
The following principles guide how MyFoodFit is being built:
- Explainability before model tuning. Logic is kept interpretable even when this limits sophistication.
- Conservative defaults where data is incomplete. The system signals uncertainty rather than inferring missing information.
- User context and preference awareness. Outputs reflect stated dietary goals and constraints, not universal recommendations.
- Transparency over prediction. The product describes what is known about a food, not what might happen if consumed.
- Iterative refinement informed by critique. Rules and thresholds are revised in response to expert review and user feedback.
These principles reflect the constraints and responsibilities of an early-stage product operating in a domain where clarity and caution are more appropriate than confidence.
Current status
MyFoodFit is in beta. We have not completed clinical validation and do not claim health outcomes. We are actively seeking collaborators to review the approach and refine the research agenda.