This document is generated on request from the responder's signed data registries — the dataset list and the reproducibility pins below are read live, not transcribed, so this dossier cannot drift from what actually produced a verdict.
The live source registry at emem.dev is unreachable right now, so the exact dataset list can't be shown. This is an honest gap, not a fabricated table — reload to fetch it, or query GET https://emem.dev/v1/sources directly. The regulation basis above is unaffected.
The live manifest registry is unreachable right now — reload, or query GET https://emem.dev/v1/manifests.
A statement is checkable three ways, none of which trust eudr.dev. See /verify to run all three on a receipt.
signer.pubkey_b32. Offline: eudr verify receipt.json.POST emem.dev/v1/verify_receipt: BLAKE3 preimage + ed25519 + merkle inclusion proof.GET /v1/log/consistency?first=<pinned> later proves the log only ever appended, never rewrote the facts the verdict relied on.Per-value integrity: every number surfaced is echo-verified byte-for-byte against its signed fact, and each fact CID dereferences at emem.dev/v1/facts/<cid>.
The per-cell deforestation verdict runs against the JRC + Hansen forest baseline over every ~10 m cell that fits inside the plot polygon. A tree-cover-2000 reading within ±2 pp of the Art. 2(4) 10% canopy threshold is flagged borderline; sub-0.5 ha losses are suppressed by the Art. 2(4) minimum-mapping-unit floor. A forest-loss signal dated after the cut-off is treated as a liability requiring the full polygon verdict — not silently absorbed. Radar (Sentinel-1 VV drop) and embedding-change are additive cross-checks, never the sole legal basis.