Kiba — iOS Pet Food Scanner

Scan a pet food barcode, get an ingredient-level, species-specific suitability score personalized to your dog or cat. Built like Yuka, focused on pets.

Live product: kibascan.com Status: Active development — M9 (UI polish & search) in progress

---

Why this repo is interesting

• Three-layer scoring engine — Ingredient Quality + Nutritional Profile + Formulation Completeness, with species-specific rules (dog DCM pulse-load advisory, cat carb overload) and per-pet personalization (breed modifiers, health conditions, allergen dual-IQ). Category-adaptive weights (55/30/15 daily food, 65/35/0 supplemental, 100/0/0 treats). See src/services/scoring/.
• Brand-blind by architecture — the scoring engine has no awareness of brand names, by design. affiliate_links is invisible to the engine at the code level, not by policy.
• 129-decision product log — every non-trivial product or technical choice is numbered, dated, and carries rationale + rejected alternatives in DECISIONS.md. Supersession is explicit (e.g., D-013 superseded by D-137 when the DCM advisory was rewritten from a grain-free gate to positional pulse-load detection).
• Supabase RLS on every user table — auth.uid() = user_id enforced in 38 migrations. External API keys live server-side only, called through Edge Functions.
• Offline-first pantry — write operations throw PantryOfflineError when the network is down; reads degrade to empty. Cache invalidation on pet-switch uses stale-while-revalidate for sub-second UI.
• Matte Premium design system — tokens, card anatomy, typography, and anti-patterns formalized in .agent/design.md. Enforced via a token-sweeper subagent.

Tech stack

Expo SDK 55 / React Native 0.83 / TypeScript 5.9 strict / Zustand 5 / Supabase (Postgres + Auth + Storage + RLS + pg_cron) / RevenueCat / Jest via jest-expo.

Architecture at a glance

  Camera scan → UPC lookup → products table
                                 ↓
                       3-layer scoring engine
                    (bypasses → Base → Species → Personalization)
                                 ↓
              ┌──────────────────┼──────────────────┐
              ↓                  ↓                  ↓
         scan_history     pet_product_scores   Kiba Index votes
         (per-user log)       (cache)        (community taste / tummy)

  Pantry → pantry_items + pantry_pet_assignments
             ↓
       pg_cron (auto-deplete every 30m)
             ↓
       Expo Push API (low stock, empty, recall, feeding, appointment)

Full walkthrough: docs/ARCHITECTURE.md.

If you are evaluating this codebase, start here

Path	Why
src/services/scoring/	The core product. 3 layers + bypasses, pure functions, deterministic.
DECISIONS.md	130 decisions with rationale, dated, superseded ones marked.
docs/references/scoring-rules.md	Authoritative scoring math: AAFCO thresholds, DMB, curves.
.agent/design.md	Design system. Tokens, card anatomy, anti-patterns.
supabase/migrations/	38 migrations. RLS applied per user-data table.
__tests__/services/scoring/regressionAnchors.test.ts	Regression anchors — Pure Balance = 61, Temptations = 0.
docs/ARCHITECTURE.md	One-page tour of layout, pipeline, and "where to look."

Running locally

This repo is primarily for code review. A full bootable demo requires Supabase + RevenueCat accounts not included here. What does work cold:

npm install
npm test            # 1508 passing / 64 suites
npx tsc --noEmit    # clean in src/ + __tests__/

To boot the app you need a Supabase project (schema in supabase/migrations/) and a RevenueCat account. Copy .env.example to .env and fill in the keys.

Regression anchors

Deterministic scoring is the product. Two canonical anchors are hard-asserted on every scoring change:

Input	Expected score	Asserted in
Pure Balance (Dog, daily food)	61	__tests__/services/scoring/regressionAnchors.test.ts
Temptations (Cat Treat)	0	__tests__/services/scoring/regressionAnchors.test.ts

Condition-modifier behavior (e.g., Pure Balance + cardiac dog → 0 via DCM zero-out, + pancreatitis → 53 via fat penalty) is covered by conditionScoring.test.ts and HealthConditionAdvisories.test.ts.

Data & IP

The product catalog is populated from publicly available retailer product pages. Only derived aggregate ingredient metadata (canonical names, severity tags, occurrence counts) is included in this repo — raw product datasets are gitignored and stay out of source control. Pipeline code is data-source-agnostic on its face.

License

LICENSE — source-available for portfolio review only. All rights reserved. No fork / redistribute / commercial use.

---

Author: Steven Diaz (steven.diaz08@gmail.com) Note: This repository is temporarily public for portfolio review (April 2026). It will be switched back to private after the review window.