LegalAutomated Decisions & AI Transparency
Automated Decisions & AI Transparency
1. Purpose
This register explains what Being Optimal automates, what inputs are used, what effect the output has, and how humans stay in the loop. It supports transparency under data-protection laws (including GDPR / UK GDPR Article 22 concepts) and product honesty for a pregnancy-nutrition planning app.
Capitalized terms follow Definitions. This document is incorporated by reference into the Privacy Policy and aligns with the Medical Disclaimer.
2. Legal posture (GDPR Art. 22 and similar)
2.1 Informational outputs, not solely automated legal/medical decisions
Under GDPR Article 22 (and UK GDPR equivalents), individuals have protections regarding decisions based solely on automated processing that produce legal effects or similarly significant effects.
Being Optimal’s posture:
| Point | Statement |
|---|---|
| Nature of outputs | Scores, flags, optimizer suggestions, lab comparisons, Knowledge Score, and AI replies are informational / educational / planning tools. |
| Medical effect | They are not clinical decisions, prescriptions, triage dispositions, or treatment orders. See Medical Disclaimer. |
| Legal effect | They do not, by themselves, grant or deny statutory benefits, credit, insurance coverage, or employment. |
| Human involvement | You choose whether to log data, accept optimizer proposals, confirm AI-proposed writes, and act on flags. Your Clinician makes medical decisions. |
| Solely automated “significant” decision | We do not treat Day Score, nutrient flags, optimizer results, blood-marker displays, Knowledge Score, or AI chat as solely automated decisions with legal or similarly significant effects within the meaning of Art. 22. |
Counsel should re-validate this posture for each launch market. If a future feature blocks access to essential services or makes binding eligibility decisions purely by algorithm, this register and the Privacy Policy must be updated before that feature ships.
2.2 Special-category health data
Automated outputs often use Special Category Health Data (pregnancy timing, meals, labs, symptoms, etc.). Processing bases, retention, and international transfers (including to AI Model Providers) are governed by the Privacy Policy and Sub Processor List — not repeated in full here.
2.3 Rights to contest, express a view, and obtain human review
Without prejudice to mandatory law, you may:
- Contest an automated output’s fairness, accuracy, or use (for example, a Grade you believe is wrong because of bad food data);
- Express your point of view about how profiling or scoring should apply to you;
- Request explanation of the logic at the level described in this register;
- Ask for human review of how *Being Optimal* processes your Personal Data in connection with these systems (product/privacy team — not a substitute for clinical care).
Contact: [email protected] Procedure: Data Subject Request Procedure
Important: “Human review” by Being Optimal means review of our data processing and product behavior. It does not mean a licensed clinician employed by Being Optimal will second-guess your obstetric care. Medical human review is your Clinician.
3. Inventory overview
| ID | Name | Logic type | Medical decision? | Primary human review |
|---|---|---|---|---|
| AD-1 | Day Score / Grade (A–F) | Deterministic engine scoring | No | You + Clinician |
| AD-2 | Nutrient shortfall / over-limit + pregnancy-critical flags | Rules-based thresholds | No | You + Clinician |
| AD-3 | Diet optimizer (MILP) | Mathematical optimization | No | You accept/reject suggestions |
| AD-4 | Blood-marker interpretation vs targets | Comparison to configured ranges | No | Clinician (diagnosis) |
| AD-5 | Knowledge Score | Quiz / learning scoring rules | No | You |
| AD-6 | AI assistant (generative) | Generative models via AI Model Providers; propose-only writes; red-flag safety redirect | No | You + Clinician; safety pre-check |
All entries: medicalDecision: false in Legal Manifest.
4. Detailed entries
AD-1 — Day Score / Grade (A–F)
| Field | Detail |
|---|---|
| What it is | An informational Day Score and letter Grade summarizing how a committed logged day aligns with nutrient adequacy models, ratios, and planning targets for your Environment (including gestational context where configured). |
| Inputs | Meal log (ingredients and quantities); nutrient totals from the catalog/engine; personal/planning targets; gestational context (e.g. trimester / GA where set); model configuration. |
| Logic | Deterministic compute in the nutrition engine (not a generative model). Same inputs yield the same score for a given model version. |
| Effect | Display of score/grade in the dashboard, insights, and related surfaces; may influence educational tips or focus areas. No automatic clinical action. |
| Human review | You decide whether the log is complete and whether to change meals. Your Clinician decides medical significance. |
| Limitations | Incomplete logs, wrong food basis (cooked vs dry), missing supplements, and model assumptions can distort grades. A high grade is not a clean bill of health. |
AD-2 — Nutrient shortfall / over-limit drivers + pregnancy-critical flags
| Field | Detail |
|---|---|
| What it is | Rules-based identification of nutrients below planning targets (shortfalls) or above planning/safety headroom (over-limits), including pregnancy-critical flags where modeled (examples: retinol/preformed vitamin A, mercury, caffeine-related limits). |
| Inputs | Nutrient totals; targets; upper-limit / safety parameters; gestational or profile context where used. |
| Logic | Threshold and ranking rules in the engine (“drivers,” fix suggestions, safety headroom). |
| Effect | Informational UI flags, explanations, and suggested focus nutrients/foods. Not a toxicity or deficiency diagnosis. |
| Human review | You interpret flags in life context; Clinician evaluates clinical risk and labs. |
| Limitations | Catalog accuracy, bioavailability simplifications, and guideline choice affect flags. National guidance may differ from product defaults. See Medical Disclaimer §5 and Editorial Content Standards. |
AD-3 — Diet optimizer (MILP)
| Field | Detail |
|---|---|
| What it is | A diet optimizer that searches for meal/quantity adjustments meeting objective modes (e.g. maximize Day Score, minimize volume subject to a score floor, DHA-forward with mercury constraints — modes as shipped in product). |
| Inputs | Current meals; targets; constraints (including user preferences/dislikes where enforced); catalog foods; safety caps where encoded (e.g. mercury). |
| Logic | Mathematical optimization (mixed-integer / linear programming style) on Cloud Infrastructure / engine pathways — not free-form generative text. |
| Effect | Suggested meal adjustments. Application requires user acceptance (you choose what to keep). |
| Human review | You accept, reject, or edit suggestions. Clinician oversees medical diets. |
| Limitations | Optimizer does not know full medical history, all allergens, cultural requirements, budget, or Clinician orders unless reflected in constraints you set. Suggestions can be impractical or undesirable. |
AD-4 — Blood-marker interpretation vs targets
| Field | Detail |
|---|---|
| What it is | Comparison of user-entered laboratory values to product-configured target or reference ranges for display and planning context (and related educational cross-checks where offered). |
| Inputs | Marker values you enter; units; dates; target ranges in product configuration; optionally linked nutrient intake context. |
| Logic | Deterministic comparison / banding against targets — not a licensed laboratory information system and not clinician chart review. |
| Effect | Informational display (e.g. below/within/above target styling). May support discussion prep with your care team. |
| Human review | Your Clinician interprets labs in full clinical context. Being Optimal staff do not diagnose from your entries. |
| Limitations | Assay methods, trimester-specific ranges, inflammation, supplementation timing, and comorbidities may not be modeled. Wrong unit entry causes wrong banding. Not diagnosis. |
AD-5 — Knowledge Score
| Field | Detail |
|---|---|
| What it is | An educational progress indicator derived from quiz / challenge / learning activity progress. |
| Inputs | Quiz progress and answers; challenge completion rules as implemented. |
| Logic | Scoring rules (points, completion, ladders — as shipped). |
| Effect | Display of educational progress; may unlock learning UX. Not a credential or clinical competency certificate. |
| Human review | You. No medical decision. |
| Limitations | Measures engagement with product education, not real-world health outcomes. Export of quiz/Knowledge Score data may be limited today — see Privacy / DSAR caveats in PREFLIGHT DECISIONS. |
AD-6 — AI assistant (generative)
| Field | Detail |
|---|---|
| What it is | In-product AI assistant (product name: Borg) that holds multi-turn conversations about nutrition planning, logging, and education. |
| Inputs | Your messages; optional health/context snippets assembled from your Environment (e.g. meal/summary context, preferences); tool results (catalog search, drafts); system instructions and safety policy. |
| Logic | Generative large-language models hosted by AI Model Providers. Additional deterministic layers may run: safety pre-check on user text, proposal safety scoring, output safety audit, nutrition-disclaimer append for dosing-like language, and propose-only write tools (user must confirm commits). |
| Effect | Conversational guidance; draft meal edits; proposed journal/lab/symptom entries; summaries. Not medical advice or emergency care. |
| Human review | You confirm every proposed write. Your Clinician for medical decisions. Automated red-flag safety redirect refuses nutrition coaching for certain urgent patterns (e.g. chest pain, heavy bleeding, can’t breathe, loss of consciousness, seizure, suicidal ideation / self-harm, eating-disorder crisis language) and points you to clinician or emergency services. |
| Data sent off-platform | Prompt and context content may be transmitted to AI Model Providers for inference. Conversation content may be persisted in our Database Provider systems while your account remains active (subject to product deletion tools and Privacy Policy retention). External provider copies follow provider DPAs; we do not currently guarantee provider-side deletion on account erasure — see Privacy Policy and PREFLIGHT DECISIONS. |
| Limitations | Hallucinations; outdated general knowledge; incomplete retrieval; imperfect red-flag detection; latency/outages. Safety layers reduce risk but cannot eliminate it. |
Canonical escalation spirit (product/community): *This one's beyond nutrition — please contact your provider or local urgent care now.*
AI urgent redirect spirit: *I'm not able to provide nutrition guidance for what you've described. Please contact your clinician or emergency services if this is urgent.*
5. What we do not automate (clarifications)
For transparency, Being Optimal does not currently use automated systems to:
- Issue prescriptions or clinical referrals;
- Provide emergency dispatch or crisis-line routing beyond static redirect text;
- Make credit, insurance underwriting, or employment decisions;
- Sell access to your identifiable health profile to advertisers (see Privacy Policy; no third-party ad pixels in the current posture described in the suite);
- Silently apply AI-proposed meal/lab writes without your confirmation (propose-only design).
If this list changes, this register will be updated.
6. Profiling
Building a nutrition Environment (meals, targets, gestational timing, optional labs/symptoms) and deriving scores/flags constitutes profiling in the data-protection sense: automated processing of personal data to evaluate aspects of your diet relative to models.
- Purpose: provide the planning product you requested.
- Consequence: personalized displays and suggestions inside the app — not third-party automated rejection of legal rights.
- Controls: you can edit/delete logs (subject to product tools), adjust preferences/constraints, stop using features, and exercise rights via [email protected].
7. Model and engine versioning
- Engine scores/flags/optimizer (AD-1–AD-4): behavior depends on engine and catalog versions. Improvements may change scores for the same log.
- Knowledge Score (AD-5): depends on quiz content versioning.
- AI (AD-6): underlying AI Model Providers and model IDs may change for quality, safety, or cost; system prompts and tools evolve. We do not guarantee a fixed model identity in the consumer UI.
Material changes to purposes of AI processing or new AI Model Providers with higher transfer risk are intended to follow notice / re-acceptance rules in Versioning And Acceptance and the Privacy Policy.
8. Accuracy and complaints
- Automated outputs can be wrong. Prefer primary food labels, lab reports, and Clinician advice when stakes are high.
- Product bugs: [email protected]
- Privacy / Art. 22-style contest and data rights: [email protected]
- Legal: [email protected]
You may also have the right to lodge a complaint with a supervisory authority in your place of residence; see regional annexes in the Privacy Policy.
9. Related documents
| Document | Why |
|---|---|
| Medical Disclaimer | Non-clinical character of all AD-* outputs; emergencies |
| Privacy Policy | Legal bases, transfers, retention, AI data flows |
| Sub Processor List | Named AI Model Providers and infrastructure roles |
| Editorial Content Standards | Nutrient data and educational evidence standards |
| Acceptable Use Policy | Misuse of AI and automation |
| Data Subject Request Procedure | How to exercise rights |
| Records Of Processing | Art. 30 accountability view |
| Definitions | Day Score, Borg, Automated Decision, AI Model Providers |
| Legal Manifest | Machine-readable AD-1–AD-6 |
10. Changes
Updates to this register follow suite versioning in Versioning And Acceptance. The version and effective date appear at the top of this document.
*Operator draft — session legal-suite-20260713. Not legal advice. Counsel review required before publication.*