Methodology
How this app generates its recommendations, what it's actually based on, and where its limits are. We'd rather tell you what we don't know than pretend otherwise.
1. Identification — you tell us what it is
There is no AI in this app. You type the item, and a plain keyword matcher (src/lib/scenarioEngine.ts) maps it to a material category. If it doesn’t match a category we have sourced data for, the app says so and gives no number instead of guessing.
The app asks targeted follow-up questions for certain categories (e.g. “is this greasy?” for cardboard) because contamination changes the answer.
2. The Three-Future Scenario Engine
The Reuse / Recycle / Dispose comparison is not generated by an AI model per request. It's a small, deterministic rules engine (src/lib/scenarioEngine.ts) that maps an item's category to a fixed lookup table of hand-authored guidance and, where available, a sourced emissions figure. The classification step (which material bucket an item falls into) is plain keyword matching, not an LLM call — which makes it testable and auditable rather than a black box. It currently has 50+ automated unit tests covering classification edge cases, contamination adjustments, and data-honesty guarantees (e.g. that reuse never claims a sourced number it doesn't have).
3. Environmental impact figures
Every quantitative figure on this site comes from the US EPA's WARM (Waste Reduction Model) documentation — pulled directly from EPA's published chapters while building this app, not estimated or invented. Each figure is net GHG emissions in MTCO2E per short ton for a given material under a given end-of-life pathway, relative to a landfilling-virgin-production baseline. A negative number means that pathway is a net emissions reduction.
| Material | Recycle/Compost | Landfill | Source |
|---|---|---|---|
| Aluminum (cans/containers) | -9.13 | 0.02 | Exhibit 2-7: Net Emissions for Metals (Aluminum Cans) |
| Glass | -0.28 | 0.02 | Exhibit 1-4: Net Emissions for Glass |
| PET plastic (#1, bottles) | -1.13 | 0.04 | Exhibit 4: Net Emissions for Plastics — PET |
| HDPE plastic (#2, jugs/tubs) | -0.88 | 0.04 | Exhibit 4: Net Emissions for Plastics — HDPE |
| Corrugated cardboard | -3.12 | 0.45 | Exhibit 8: Net Emissions for Paper Products — Corrugated Containers |
| Mixed paper (office paper, mail, etc.) | -3.53 | 0.4 | Exhibit 8: Net Emissions for Paper Products — Mixed Paper (general) |
| Food scraps / organic waste | -0.15 | 0.5 | Exhibit 1-10: Net Emissions for Food Waste |
We deliberately do not have a figure for textiles in this build — the relevant WARM documentation chapter wasn't reachable while researching this, and we'd rather say so than guess. Reuse is also never assigned a quantitative figure for any material, because WARM doesn't model reuse as an end-of-life pathway at all — it only covers recycling, composting, combustion, and landfilling.
4. Local Reality Check — an honest limitation
We do not have a live integration with any municipal recycling-rules database or waste-hauler API in this build — no free, self-serve option exists for that (the established ones, like Earth911 or Recycle Coach, are partner/B2B programs we don't have access to). What the app provides instead is clearly labeled a plain note that rules vary by city, plus a link to a local lookup (Earth911).
Adding verified city data later would mean adding a data source behind this card; none is claimed today.
5. Known limitations
- Typed items are matched by keywords; unusual items may not be recognized, and the app says so.
- Impact figures are national-average, category-level estimates, not measurements of your specific item.
- Local guidance is general, not verified against your actual municipal program.
- Textile emissions data is currently missing (see above) rather than guessed.
- The "surprising fact" on each Learning Receipt is derived from the same sourced figures shown on that page — it isn't a separate, unverified claim.