How we keep the community safe

Babylore lets parents share recipes from their own kitchens with parents around the world. The whole reason this works is that the recipes are real — your grandmother's pastina, her mother's miyeokguk, a Saskatoon-apple purée your aunt makes in summer. Real food, real culture, real authenticity.

But baby food is the one category where the wrong recipe can actually hurt a child. Honey before 12 months can cause infant botulism. Whole grapes are a choking hazard for toddlers. Raw fish carries parasites a baby's immune system can't handle. We can't just let anyone upload anything and call it a feed.

So every recipe submitted to the community runs through three layers of safety review before it reaches another parent's screen. Here's how they work — in plain English.

Layer 1 — Deterministic rules

The first layer is a set of explicit, well-documented rules for foods that are dangerous at certain ages. When you submit a recipe, we check the ingredients and the minimum age you've set against these rules. If any rule fires, the submission is rejected before it's ever saved — with a specific reason so you can fix it and try again.

The current rules cover, among others:

Honey under 12 months. Infant botulism risk. Hard block.
Raw fish (sashimi, ceviche, tartare, etc.) under 24 months. Parasite and heavy-metal risk.
Raw or undercooked eggs under 12 months. Salmonella risk.
Whole nuts under 36 months. Choking hazard. Chopped, ground, or smooth nut butter is fine.
Whole grapes or cherry tomatoes under 36 months. Unless halved or quartered — they're the most common toddler choking hazard.
Cow's milk as a drink under 12 months. Recommended against by pediatric guidance.
Serving raw or undercooked poultry, fish, or meat at any age. Foodborne illness risk.
Recipes targeting babies under 4 months. Solid foods don't generally start before then.

This list grows. If pediatric guidance updates or community feedback surfaces a new pattern worth blocking, we add the rule and every future submission is screened against it.

Layer 2 — AI-assisted nuanced review

The deterministic rules catch the well-documented don'ts. But plenty of safety concerns are subtler — a recipe with thick peanut butter for a 7-month-old (choking texture), an omelette that doesn't mention cooking through, sugar in a 6-month-old purée, allergens left untagged. Rules can't easily encode these because the concern lives in context, not in a literal word match.

For these, we use a second layer: an AI review running on Anthropic's Claude API with a pediatric-nutrition prompt. When a submission passes Layer 1, we send the title, ingredients, and steps (not your name, not your email, not any other identifier) to the model and ask: are there concerns at the stated age — choking hazards from preparation, allergens that weren't tagged, dangerous techniques in the steps, or a minimum age that seems too low for what's described?

If the model returns "safe," the recipe auto-publishes and appears in the community feed within seconds. If it flags concerns, the recipe is held and surfaced to a human for review. The model's reasoning is annotated so the human can quickly understand what to look at.

Two things worth being clear about:

The AI does not train on your recipe. Anthropic's API explicitly does not use the data we send for training.
The AI is not a doctor. It's a second pair of eyes on top of the rules. Some concerns will still slip through. The human-review and community-report layers exist to catch what the AI misses.

Layer 3 — Community reports

Once a recipe is live, every signed-in user can flag it for review. Reports cover six reasons: unsafe content, wrong minimum age, missing allergen tag, inappropriate content, spam, and a free-form "other." If the same recipe gets three unresolved reports from different users, it's automatically unpublished and goes back to the moderation queue for human review.

One report per user per recipe — so a single bad actor can't rage-flag content into oblivion. The threshold of three is deliberately low. False positives are easy to fix (re-publish after review). False negatives (real hazards staying live) aren't.

What stays your job

The safety pipeline is meaningful but not infallible. No automated system catches every concern, and every baby is different. A few things we ask you to keep doing as a parent:

Read every recipe before you make it. The ingredients, the steps, the minimum age. Trust your eyes, especially on community submissions.
Adjust to your baby's stage. A 9-month-old's texture handling isn't the same as a 12-month-old's. Mash, chop, or thin as needed.
Know your allergens. If your baby has a known allergy or a family history, double-check every ingredient.
Talk to your pediatrician. About introducing solids, allergens, and any food you're unsure about. Babylore doesn't replace that conversation — it gives you something concrete to bring to it.
Supervise every feeding. Sit your baby upright. Stay in the room. Watch every bite, especially for new textures.

What you can do if something looks wrong

If you see a recipe in the community feed that looks unsafe, mislabeled, or otherwise problematic:

Tap the report button on the recipe.
Or email help@suyu.ca directly. A real human reads every message.

We review every report. Recipes confirmed unsafe come down, their authors are notified, and repeat offenders lose their ability to publish. Reports that turn out to be a misunderstanding (texture preference, cultural variation, minimum-age difference between countries) get a written response explaining why we kept the recipe up.

What we won't do

We won't outsource moderation to no-one. A pipeline that exists only on paper isn't a pipeline. Every layer above is live in production.
We won't pretend the pipeline is perfect. It isn't. It's a strong first line, paired with your judgment as the parent who's actually feeding your child.
We won't share recipe content with advertisers, marketing partners, or data brokers. The only third parties who see recipe text are Anthropic (for the AI review) and Supabase (for hosting). Both are described in detail in our Privacy Policy.

For the legal details

Our Recipe & content disclaimer covers the formal terms around bundled and community recipes, allergen guidance, choking hazards, the limits of automated review, and our policy on reporting concerns. The Terms of Service and Privacy Policy cover the broader legal frame.

Contact

Questions or recipe safety concerns: help@suyu.ca.