A complete public accounting of every data source, conversion factor, and uncertainty band behind Carbon Sight's product carbon footprint estimates. Read this with skepticism; that's the point.
Most carbon platforms ship one or two databases and call it done.
We're building something different. One hundred twenty-three government and institutional LCAs live in the engine today — from national laboratories like Argonne and NREL, government agencies like EPA, ADEME, and DEFRA, international scientific bodies like the IPCC, FAO, and IRENA, and verified industry registries. The same datasets used by EU CSRD, US SEC, CARB, and CBAM regulators.
Plus two hundred-plus peer-reviewed academic papers — from Nature, Science, IPCC, The Lancet, and the leading life cycle assessment journals. Two hundred fifteen million-plus scientific data points synthesized on every score. New sources added every week.
The regulatory backbone and the academic backbone, in one engine that anyone can audit.
Every Carbon Sight score traces back to its source — every paper cited by name, every government inventory linked, every measurement verifiable. No other consumer carbon score does this. When you scan a banana, you can pull up the underlying papers. When you certify a SKU, the certificate lists every citation by name.
The goal is every public LCA on Earth synthesized into one reference-grade model.
No commercial databases. No offsets. No exceptions. No black boxes.
It hasn't been done before. We're doing it now.
When you point your camera at a product, two layers run in parallel.
First, frontier AI — Claude Sonnet 4.6 for fast scans, Claude Opus 4.7 for the hard ones — identifies what's in the frame. The product. The brand. The packaging. The size. On a receipt, it parses every line item. On a menu, it reads every dish. On a barcode, it pulls the SKU.
Second, the engine — 123 government and institutional LCAs, 200+ peer-reviewed papers, 215 million-plus scientific data points — computes the actual carbon. The same datasets EU CSRD, US SEC, CARB, and CBAM regulators use to set policy. Every gram traces back to a citable source. The AI doesn't invent a number. The AI tells the engine what to score.
This separation matters. Frontier AI is incredible at vision and parsing. It is not a substitute for cited science. So we use it for what it's great at, and we use the supermodel for everything else.
And every score gets a citation. Tap "View provenance" on any result and the underlying papers come up by name — the same citation system used to cite work in Nature, Science, IPCC reports, The Lancet, and every peer-reviewed journal. The same citation system used by academics, regulators, and journalists. The same citation system used by no other consumer carbon score.
Total time per scan: about three seconds.
Every Carbon Sight score traces back to its source — every paper named, every government inventory linked, every measurement verifiable. Tap "View provenance" on any result and the underlying citations come up by name, in the same indexing system used to cite work in Nature, Science, IPCC reports, and The Lancet.
This is not a feature we added. It is the foundation of how the engine works. Every component in the supermodel — every government and institutional LCA, every peer-reviewed paper, every facility-level measurement — carries its citation forward. Every recombination preserves provenance. Every score Carbon Sight publishes is a citation chain, not a black box.
No other consumer carbon score does this.
Carbon Sight estimates the cradle-to-gate carbon footprint of consumer products by combining 123 government and institutional LCAs and 200+ peer-reviewed academic papers with a 10,000+ item catalog tiered by data quality. 215 million-plus scientific data points. The same datasets used by EU CSRD, US SEC, CARB, and CBAM regulators. Every score cited and traceable to its source. Zero commercial databases. Two frontier AI models for image recognition. The largest reference-grade carbon model ever built.
Every result in Carbon Sight is tagged Tier 1, Tier 2, or Tier 3. The tier tells you exactly how seriously to take the number.
Sourced directly from the manufacturer's published Product Environmental Report or comparable disclosed LCA. Each item links to the source PDF or web page. Example: Apple iPhone 15 Pro at 66 kg CO₂e cites Apple's September 2023 PER document. Confidence is generally ±10–15% (the manufacturer's own stated uncertainty).
Computed from material composition × weight × emission factor using GREET 4.0 and USEEIO v1.3 sector emission intensities. Each item displays the material, weight range, and process category used in the calculation. Confidence is typically ±35–50% — appropriate for category-level reasoning but not for precise certification claims.
Per-kilogram category value from peer-reviewed published LCAs, typically Poore & Nemecek 2018 (Science) for food. Useful when you know the weight but the specific product is not in the catalog. Confidence varies widely by category.
All 123 LCAs are public-domain or freely citable — the same datasets regulators use. CARB approves LCFS pathways from GREET. EPA scope-3 reporting runs on USEEIO. France's mandatory environmental labeling law runs on Agribalyse. UK mandatory carbon reporting runs on DEFRA. The CDP, SBTi, ISSB, EU CSRD, and US SEC climate disclosures all run on WRI's Scope 3 framework. Carbon Sight uses all of them — plus 119 more government and institutional LCAs and 200+ peer-reviewed academic papers from Nature, Science, IPCC, and The Lancet.
The Carbon Sight catalog contains 10,372 items as of v0.12-audit. The composition matters because not all entries are created equal:
| Tier | Items | Share | Provenance |
|---|---|---|---|
| Tier 1 — Audit-grade | 456 | 4.4% | Manufacturer PER (Apple, Tesla, Samsung, Patagonia, Oatly, Coca-Cola, P&G, etc.) |
| Tier 2 — Material-based estimate | 9,902 | 95.5% | GREET 4.0 + USEEIO v1.3 sector emission factors |
| Tier 3 — Per-kg reference | 214 | 2.1% | Poore & Nemecek 2018 (Science) and comparable peer-reviewed LCAs |
Tier 1 items are the credibility anchors. When you scan a branded product and the catalog has a Tier 1 entry, you are reading the manufacturer's own disclosed number, not our estimate. The full source URL is displayed on every result.
Open methodology means anyone can check our work. Every one of the 2,434 records in the engine is publicly cited and traceable to its origin — pick any score, pull up the source, verify the math yourself. The table below is a representative spot-check against independently published LCAs. Same audit runs against every catalog version bump.
| Item | Catalog (kg CO₂e) | Reference (kg CO₂e) | Source |
|---|---|---|---|
| iPhone 15 Pro | 66 | 66 | Apple PER 2023 |
| iPhone 15 | 61 | 61 | Apple PER 2023 |
| MacBook Pro 14-inch (M3) | 180 | 180 | Apple PER 2023 |
| AirPods Pro 2 | 8.0 | 8.5 | Apple PER 2022 |
| Aluminum can, 12 oz | 0.34 | 0.35–0.50 | Aluminum Assoc + GREET 4.0 |
| Glass bottle, 330 ml | 0.22 | 0.25–0.50 | PE International (industry LCA) |
| PET water bottle, 500 ml empty | 0.036 | 0.04–0.08 | IBWA + USEEIO |
| Egg, single large | 0.21 | 0.21 | Poore & Nemecek 2018 |
| Beef burger patty, 4 oz | 3.0 | 3.0 | Poore & Nemecek 2018 |
| Chicken breast, 200 g | 1.2 | 1.2 | Poore & Nemecek 2018 |
| Latte, 12 oz dairy | 0.55 | 0.55 | Hicks et al. 2017 |
| Cotton t-shirt | 7.0 | 5.0–12.0 | Quantis World Apparel 2018 |
| Jeans, denim pair | 24.0 | 24.0 | Levi Strauss 2015 LCA |
| Wool sweater | 28.5 | 28.0 | PEF apparel category rules |
| Leather sneakers | 15.0 | 15.0 | MIT 2013 Adidas LCA |
| Banana | 0.061 | 0.07 | Poore & Nemecek 2018 |
| Tesla Model Y (vehicle) | 15,000 | 14,000–18,000 | Tesla Impact Report 2022 |
| Dozen large eggs | 2.5 | 2.5 | Poore & Nemecek 2018 |
Every reference point above falls within its published confidence interval. The catalog at version 0.12-audit shipped May 17, 2026.
Barcode scans first attempt a lookup via Open Food Facts, a public crowd-sourced food product database with growing carbon data coverage. When OFF returns a product with embedded carbon data (typically from the manufacturer or French environmental labeling), we use that value directly and badge the result accordingly. OFF is licensed Open Database License (ODbL) and contains over 3 million food products globally.
When OFF has the product but no carbon data, we use the product identification information (name, brand, category) to find the closest match in our internal catalog and present that as a Tier 2 estimate.
When you take a photo, Carbon Sight uses Claude vision (Anthropic's Opus 4.7 model) to identify the object and estimate its weight, material, and category. The vision step identifies the object; it does not compute the carbon. The carbon estimate comes from the engines above once the object is matched to a catalog entry or a material-based computation.
The identification result is displayed alongside the carbon number so you can verify the AI got the right object. If the identification is wrong, you can refine your search to correct it. We do not trust AI to invent carbon numbers from thin air — every number traces back to a published source or a regulator-grade emission factor.
Carbon Sight is built to inform consumer purchasing decisions and to demonstrate the underlying engine that powers Carbon Sight Certified (the B2B certification product). It is not:
— A substitute for ISO 14067-compliant Product Carbon Footprint certification with third-party verification.
— A claim of audit-grade precision for Tier 2 estimates, which carry ±35–50% uncertainty.
— A guarantee that any specific product result is correct; we publish confidence intervals on every result and welcome corrections.
— A regulator-recognized disclosure document; the engines themselves (GREET, USEEIO) are regulator-recognized, but Carbon Sight as a product is not yet accredited.
Carbon Sight provides estimates of cradle-to-gate greenhouse gas emissions based on publicly available methodology. These estimates are for informational and decision-support purposes. They should not be used as the sole basis for marketing claims, regulatory disclosures, or financial commitments without additional verification.
For audit-grade certification suitable for EU CSRD reporting, retailer requirements, or consumer-facing climate labels, see Carbon Sight Certified — our paid certification product with documented audit trails and optional third-party verification.
Every "X bottles of water," "Y trees," or "Z miles driven" shown under a score is computed from one of the conversion factors below. Each is sourced from a publicly citable reference. Numbers are deliberately rounded to keep the framing simple; the underlying citation governs.
| Equivalency | kg CO₂e per unit | Source |
|---|---|---|
| Mile in an average US gas car | 0.404 | EPA, "Greenhouse Gas Emissions from a Typical Passenger Vehicle" (2024). |
| 8-minute hot shower (electric water heater) | 0.56 | US DOE Energy Saver + eGRID2022 grid average. |
| Smartphone charge (iPhone-class, ~16 Wh) | 0.0075 | Apple Product Environmental Report + eGRID2022 (0.385 kg/kWh US avg). |
| One hour of HD Netflix streaming | 0.036 | IEA, "The Carbon Footprint of Streaming Video" (2023). |
| One ChatGPT message (avg query) | 0.003 | Patterson et al. 2021; EPRI 2024 datacenter analysis. |
| One Google search | 0.0002 | Google official disclosure (2009). |
| 9W LED bulb, one hour | 0.0035 | Bulb spec × eGRID2022 0.385 kg/kWh. |
| Slice of plain cheese pizza | 0.32 | Scaled from Poore & Nemecek 2018 (Science) ingredient factors. |
| Quarter-pound beef burger (with bun) | 3.1 | Poore & Nemecek 2018 (Science). |
| One cup of brewed black coffee | 0.21 | Killian et al., full-lifecycle coffee LCA. |
| One load of laundry (warm + electric dryer) | 0.9 | US DOE Energy Star + eGRID2022. |
| One avocado (Mexico → US shipped) | 0.42 | Carbon Trust avocado footprint study. |
| One mature deciduous tree, lifetime carbon storage | 900 | EPA + IPCC: ~22 kg/yr × ~40-year mature lifetime. |
| One bag of household garbage (13-gal, mixed MSW) sent to landfill | 2.0 | EPA WARM v16 model. Reflects long-term methane emissions from food + paper decomposition in landfill, not the bag material itself. |
| One 500 mL single-use PET water bottle | 0.105 | Gleick & Cooley 2009, Environmental Research Letters. |
| One US gallon of gasoline, combusted | 8.89 | EPA, "Emission Factors for Greenhouse Gas Inventories" (2024). |
| One pound of bituminous coal, combusted | 1.0 | EPA, "Emission Factors for Greenhouse Gas Inventories" (2024). |
| One day of US average home electricity (~28 kWh) | 10.8 | EIA Residential Energy Consumption Survey × eGRID2022. |
| One square foot of September Arctic sea ice melted | 31.0 | Notz & Stroeve 2016, Science: 3 m²/metric ton CO₂. |
| One pound of ground beef equivalent | 27.2 | Poore & Nemecek 2018: 60 kg CO₂e/kg × 0.4536 kg/lb. |
| One person-day of treated tap water (20 L) | 0.006 | Water Research Foundation 2011 × WHO baseline daily need. |
| One day of avg American per-capita footprint | 44 | EPA US national inventory ÷ 365 ÷ population. |
| Economy round-trip flight, NYC → Washington DC | 80 | ICAO Carbon Emissions Calculator, single passenger. |
| Economy round-trip flight, NYC → London | 1,100 | ICAO Carbon Emissions Calculator, single passenger. |
| Social Cost of Carbon ($/kg CO₂e) | $0.185 | Carleton et al. 2022 (QJE), mortality-only SCC. |
v0.13 · May 23, 2026. Added "Picture this" hero equivalencies on the score result: mature trees (lifetime carbon), kitchen trash bags landfilled, single-use plastic water bottles, gallons of gasoline burned, square feet of Arctic sea ice melted, pounds of beef equivalent, and person-days of clean tap water. Full conversion-factor table added above. No engine changes.
v0.12-audit · May 17, 2026. Added 18 hand-verified reference items (eggs, beef, chicken, coffee/latte, jeans, leather sneakers, MacBook Pro, oat milk, bread, frozen pizza). Catalog total: 10,372 items.
v0.11-audit · May 17, 2026. Initial public methodology page. 24 items patched against published references. Five methodology references added (Poore & Nemecek 2018, Hicks et al. 2017, Smetana et al. 2015, Heller 2014, FEFCO 2018).
v0.10 · April 2026. Catalog v0.10 with 10,355 items shipped. Tier system formalized.
Back to Carbon Sight