Team Pyragrid · IEEE ClimateChain 2026
CarbonTruth is the SSL certificate for climate data — a verification layer beneath the blockchain that turns raw satellite imagery into unforgeable, retire-once carbon credits. Not another marketplace. The trust the market is missing.
Peer-reviewed studies keep finding the same thing: most credits don't represent real carbon. The data going onto the ledger was never trustworthy in the first place.
A 2023 study in Science of 89 million credits found only ~6% backed by real, additional carbon. The other 94% are tonnes someone paid to remove — still in the air.
West et al., Science 2023 · doi:10.1126/science.ade3535A 2025 Science analysis of 52 REDD+ projects found just 19% delivered the emission reductions they claimed. Verification, not ledgers, is the bottleneck.
Tang et al., Science 2025 · doi:10.1126/science.adw4094Put an unverified credit on-chain and all you've done is make a bad number immutable and tradeable. Transparency about garbage is still garbage.
The hard part is getting a trustworthy measurement onto the chain in the first place. That's a verification problem, and it's where every registry fails.
AI-verified from independent open data, anchored with an evidence hash, made globally unique by a geo-temporal fingerprint. Verify once, trust everywhere.
Pick a real project, set the claim, submit it to the oracle network, mint it, sell it, retire it, and try to forge the ledger. Every step is a real action you take — the full use-case, running in your browser.
| VCC ID | Project | tCO₂e | Owner | Status | Actions |
|---|
Everything here is the real logic — band math, inverse-variance fusion, SHA-256 fingerprinting, the double-count and retire-once invariants, chain verification, and ownership transfer — running live in your browser, with live Sentinel-2 scene discovery through a serverless function. The NDVI→biomass step is a trained neural network (6→32→16 MLP, R²≈0.98 on held-out data) calibrated on GEDI-style labels; satellite pixels are a bundled representative scene, not yet field-validated at this plot. The identical engine runs offline as pure Python — python app.py (the net ships as frozen weights, inference is dependency-free).
Sentinel-2 red & near-infrared bands over the plot, before and after. We query Microsoft's free Planetary Computer for live scene IDs — no API key.
A trained neural network maps spectral features (NDVI, SAVI, NIR…) to aboveground biomass, then to CO₂e. A counterfactual baseline means we only ever credit carbon that's genuinely additional.
Optical, GEDI lidar and Landsat archive are fused by inverse-variance weighting. Mint only if confidence clears the gate and k-of-n oracles agree.
A geo-temporal fingerprint makes each tonne globally unique — one token per fingerprint, so double-counting can't be expressed. Retire-once, forever.
Because CarbonTruth is a verification layer, not a marketplace, the same Verified Climate Claim works across all four ClimateChain tracks — and UN Article 6.4 now mandates exactly this kind of verification.
Over-crediting caught before mint; every tonne backed by independent evidence.
The same receipt certifies clean-energy generation claims and RECs.
Verified emissions claims flow across Scope-3 boundaries without double-counting.
Any sensor-backed environmental claim gets a portable, auditable trust stamp.
Market figures are third-party projections shown for context, not revenue forecasts. CarbonTruth monetizes verification (≈ a dollar-fifty per credit at near-zero marginal cost), not the tonnes themselves.
Every claim on this page is traceable to peer-reviewed science or an open standard.
West, T. et al. (2023). Action needed to make carbon offsets from forest conservation work for climate change mitigation. Science. — Only ~6% of 89M credits backed by real carbon.
Tang, F. et al. (2025). Carbon integrity of REDD+ projects. Science. — 19% of 52 projects met their emission-reduction targets.
Chave, J. et al. (2014). Improved pantropical allometric models to estimate aboveground biomass. Global Change Biology. — The AGB equation in stage 2.
Data sources. Sentinel-2 L2A via Microsoft Planetary Computer (open STAC, no key); NASA GEDI spaceborne lidar; Landsat archive. IPCC carbon fraction CF = 0.47.
No black boxes — every layer is something a reviewer can inspect, re-run, or verify for themselves.