One app: tell it what you have → see what you can cook → get the list of what you're missing — across any cookbook you plug in. The Robot Cookbook is just source #1. This page is the project hub: the app, the specs, and the earlier prototypes it grew out of. More projects at app.memorysphere.org.
The unified flow. Tap staples or paste a receipt → recipes re-sort into make now and 1–2 items away → smart nudges → a missing-items basket you can send to Instacart. Bring your own recipes (paste / upload / schema.org) and they flow through the same engine. This is the product.
Open the app →The product + engineering plan in plain English: competitive landscape, what's defensible, the source-adapter architecture, the matching engine & its big-O/scalability, the Phase-2 backend (vision photo-parse, URL import, Instacart), real cost tables, business models, and risks. Start here to understand the whole thing.
Read the spec →The business model in plain English: free to add (virality), earn on the Instacart cart + a small premium, and why the 30¢-per-charge fee floor means you can't micro-charge.
Read the money plan →The earlier single-purpose version of the cook-from-receipt idea (now folded into the app above). Kept for reference.
Open →Standalone ingredient filter: multi-select, match-any / match-all, and an “I have these → what can I make” pantry mode. (Also folded into the app.)
Open search →An experimental force-directed map of recipes ↔ ingredients (deprioritized — the network look read as a hairball). Kept for reference.
Open →The written findings: ranked pantry staples, greedy “core pantry” set-cover tiers, the natural recipe clusters, and cook-together bundles. Markdown — open in any viewer.
Read the report →The simpler SVG version of the recipe↔ingredient network. Reference only.
Open →The main app is at cook.memorysphere.org (this repo's index.html). Everything reads the same recipes.json; the older prototype pages are kept for reference. Part of MemorySphere Labs.