Inspiration## Inspiration

Palm oil buyers now require proof that smallholder plots aren't linked to deforestation, and EUDR-style compliance is pushing this fast, but satellite monitoring is too coarse for small, mixed-crop holdings, and manual field verification is too expensive to scale to millions of smallholders. We wanted a way for a farmer to prove their own plot's status directly, using nothing but a phone, without an expensive site visit.

What it does

PlotProof will let a smallholder capture live, on-site photo evidence of their plot. The evidence will be run through a series of anti-spoofing checks (GPS plausibility, timestamp freshness, duplicate detection, and screen-replay/tamper detection using image forensics) before it's accepted. Verified evidence will be pinned to IPFS and anchored on-chain, creating a permanent, tamper-evident record that auditors or buyers can check without a costly field agent. Farmers will earn reputation points for verified submissions.

How we plan to build it

Frontend: React/Vite app with live in-browser camera capture (no file uploads, to prevent farmers submitting old or stock photos)
Backend: Node/Express service running a multi-signal anti-spoof pipeline, GPS bounding-box and accuracy checks, EXIF consistency, duplicate-hash detection, and computer-vision heuristics (glare uniformity, bezel/edge detection, moiré pattern detection, motion parallax across a frame burst) to catch photos of a screen or printout rather than a real field
Smart contract: Solidity contract on an EVM testnet to store the verified record's hash, GPS coordinates, timestamp, and IPFS CID, and credit reputation points to the submitter
Storage: IPFS (via Pinata) for the verified image, kept off-chain for cost; only the hash and CID will go on-chain

Challenges we anticipate

Distinguishing a real live capture from a photo-of-a-photo without expensive ML models is the hardest part of the design. Our plan is to combine several weak, individually-fallible signals (parallax, glare, moiré) so that no single trick fools the system, while routing ambiguous cases to manual review instead of a hard auto-reject.

What we're most excited to build

Designing an anti-spoof pipeline that can run entirely from a phone camera with no special hardware, and a verdict system that treats forensic signals as evidence to weigh rather than binary pass/fail gates.

What we expect to learn

That "verify the evidence is real" and "verify what's in the evidence" are two different problems, plausibility/liveness checks alone won't confirm land-cover status, which is why we've scoped that as future work below rather than part of the MVP.

What's next for PlotProof

Add a land-cover/vegetation classification step (satellite baseline cross-reference or an on-device model) so the system can eventually speak to deforestation status directly, not just evidence authenticity
Move from coarse regional GPS bounding boxes to real plot-level polygon boundaries
Add cryptographic attestation (backend-signed EIP-712 data) so the smart contract only accepts submissions that actually passed the anti-spoof pipeline
Pilot with a smallholder cooperative to validate the review workflow for flagged submissions