Households throw away the wrong way — because nobody tells them the honest answer.

Most waste decisions happen at home, in seconds, with no good information. Items get landfilled that could have been reused, composted or recycled — and contaminated items get tossed into recycling bins, which can spoil whole loads.

• Only about 21% of recyclable material households put out is actually recycled (Recycling Partnership, 2024) — source.
• The U.S. recycled and composted about 32% of municipal solid waste (EPA, 2018 data); roughly half was landfilled, and food was the largest single landfilled material — source.
• Landfilled food and paper emit methane. In EPA WARM, landfilling food waste is a net increase (+0.50 MTCO₂E/short ton) while composting is a net decrease (−0.15) — source.

Every household that sorts waste — especially renters, students and families without access to clear local guidance — plus the municipal recycling facilities that pay to sort contaminated loads, and communities living near landfills. Climate impact is global; the decisions that drive it are made one item at a time.

• City flyers and charts are generic: they don’t answer “what about this item?”
• Quick “is it recyclable?” answers hide that grease, coatings and local rules change the result.
• Almost none compare reuse vs. recycling vs. disposal side by side, or show how strong the evidence is.

You type an item and 3R-AI asks the questions that change the answer (is it empty? greasy?). A transparent, tested rules engine compares three futures: reuse, recycle, dispose. Emission figures come only from EPA WARM; where no sourced figure exists, the app says so instead of inventing one. A Learning Receipt turns each decision into a lesson, and an optional weight input estimates the CO₂e difference versus landfilling.

We built the data layer from EPA WARM documentation (per-entry citations are in the code), designed the rules engine to be deterministic and unit-tested (50+ automated tests), and made every recommendation traceable to a rule or a source. Bugs found by the tests (e.g. a cardboard tube misclassified as corrugated, a carpet matching “PET”) were fixed and are now regression tests. See Methodology.

Impact model: each logged decision can record an estimated kg CO₂e avoided versus landfilling = (WARM pathway − WARM landfill) × 0.5 kg per lb × weight. These are category-level national averages, not measurements. We have not measured real-world impact yet and make no claim of it.

Feasibility: runs on a free-tier stack (Next.js, Vercel, Firebase) with no hardware. Next step: a school or classroom pilot, measuring decisions logged and reflection answers. City-specific rules can plug into the existing provider interface; no municipal partnership is claimed.

To measure the problem rather than assume it, the app includes a short anonymous quiz where people choose where 12 common items go. Aggregate results (how many people, how many got each item right) are shown on the results page exactly as collected, from a small self-selected sample, so they illustrate the problem but are not a study.

• Typed items are matched by keywords, so unusual items may not be recognized; the app says so rather than guessing.
• WARM figures are national averages and exclude reuse; textiles have no sourced figure here.
• Local recycling rules vary; the app gives general guidance and says when it can’t verify local rules.