CarbonTruth

---

What this is

Peer-reviewed science found that ~94% of 89M forest-carbon credits were phantom (West et al., Science 2023) and only 19% of REDD+ projects met their targets (Tang et al., Science 2025). A blockchain can't fix that — it only proves a number didn't change, not that it was ever true. That's the oracle problem.

CarbonTruth solves the half everyone else skips. One primitive — a Verified Climate Claim (VCC) — is AI-verified from independent open satellite data, anchored on-chain with an evidence hash, and made globally unique by a geo-temporal fingerprint so a tonne can never be double-counted. One primitive maps onto all four hackathon tracks (carbon markets, renewable energy, supply chains, climate data).

Verify once. Trusted everywhere.

This repo

carbontruth/
├─ index.html            # the public site + interactive app (static, deploys on Vercel)
├─ logo.png
├─ api/
│  └─ stac.js            # serverless fn: live Sentinel-2 discovery (Planetary Computer)
├─ vercel.json           # Vercel config (static site + /api function)
├─ package.json
├─ prototype/            # the real engine — pure-stdlib Python (python app.py)
│  ├─ app.py             #   satellite → model → oracle → ledger, HTTP explorer on :8017
│  ├─ ct/                #   satellite.py · model.py · oracle.py · ledger.py
│  └─ contracts/
│     └─ CarbonTruthVCC.sol   # real EVM contract: fingerprint-uniqueness + retire-once
└─ docs/
   └─ demo-video-script.md

The website runs the entire pipeline live in the browser — NDVI band math, inverse-variance oracle fusion, SHA-256 fingerprinting, a hash-chained ledger, ownership transfer and retirement — and you drive the whole credit lifecycle yourself: pick a project, set a claim, submit it to the oracle network, mint it, trade it, retire it (download the certificate), and try to forge the ledger.

The prototype/ folder is the same logic as a zero-dependency Python service, so the demo is backed by real code, not a mock-up.

Deploy to Vercel

One-click (dashboard):

1. Push this repo to GitHub (see below).
2. On vercel.com → Add New → Project → import the carbontruth repo.
3. Framework preset: Other (no build step). Leave Build & Output empty. Click Deploy.

That's it — index.html is served at /, and api/stac.js becomes https://<you>.vercel.app/api/stac.

CLI:

npm i -g vercel
vercel          # preview
vercel --prod   # production

Run the Python engine locally

cd prototype
python app.py              # → http://127.0.0.1:8017
python app.py --selftest   # honest ✓ · duplicate blocked ✗ · inflated rejected ✗

No pip install, no API keys — Python standard library only (tested on 3.14).

How it works

Stage	File	What it does
Observe	ct/satellite.py	NDVI = (B08 − B04)/(B08 + B04); live Sentinel-2 via Planetary Computer
Estimate	ct/model.py	NDVI → biomass → CO₂e; counterfactual baseline → additional only
Verify	ct/oracle.py	inverse-variance fusion μ = Σwᵢmᵢ/Σwᵢ, mint iff `V = 1 −
Mint	ct/ledger.py	SHA-256 hash chain; fingerprint f = H(geohash‖date‖projectID‖k); retire-once R ∩ A = ∅

Evidence

• West, T. et al. (2023) Science — doi:10.1126/science.ade3535
• Tang, F. et al. (2025) Science — doi:10.1126/science.adw4094
• Chave, J. et al. (2014) Global Change Biology — doi:10.1111/gcb.12629
• IPCC carbon fraction CF = 0.47; Sentinel-2 / NASA GEDI / Landsat open data

Scope honesty: the logic is real; satellite pixels are a bundled representative scene and the NDVI→biomass relation is a transparent prototype estimator, not a field-calibrated model. Dollar and tonne figures illustrate the method. The production gap is field data and calibration — not the architecture.

---