From sensor to proof.
The Mālama dMRV cycle. Four steps from physical reality to on-chain certificate. The same pipeline serves carbon project sites and AI data centers. Hardware signs at the source. Hex Nodes validate. Cardano anchors. Proof of Truth issues.
Four steps. One chain of custody.
01 · Direct Sensor Capture
IoT sensors deployed at the source capture high-frequency environmental data. For carbon: Genesis 300 nodes at biochar and enhanced rock weathering sites measure soil temperature, moisture, EC, pH, and atmospheric conditions. Forestry, soil carbon, and regenerative agriculture support are roadmap items. For AI compute: rack-mount AI Power Sensors measure exact electrical load per inference and training cycle, plus cooling water and evaporation.
Every reading is signed by an ECDSA private key burned into the device's ATECC608B secure enclave at manufacture. The signature exists before the data leaves the silicon. There is no software path to forge it.
02 · Edge Verification
Raw signed data is processed at the edge before transmission. The verification stack runs cryptographic, protocol, physical, spatial, temporal, and methodology checks in sequence. AI z-score anomaly detection flags spoofed and physically implausible readings.
For AI compute, edge verification cross-references power draw with real-time grid carbon intensity and cooling water evaporation rates. A kilowatt-hour at 4 AM in Quebec carries different carbon weight than the same kilowatt-hour at 4 PM in Texas. Edge verification captures that difference at the moment of measurement.
03 · Hex Node Consensus
Verified data is broadcast to the Mālama Hex Node network. Each Hex Node operates within an H3 geographic cell governed by an NFT-HEX rights object. Nodes perform decentralized audits, verifying that the data packet matches protocol standards and the device signature checks out against the on-chain device registry.
Validated readings are batched by region and methodology, bundled into Merkle trees, and archived on Arweave for permanent data availability. Only the Merkle root is committed to Cardano-anchored ledger via CIP-25/CIP-68 SaveCard semantics. Validators stake MLMA. Fraudulent attestations face 10% slashing. The network is Byzantine Fault Tolerant by construction.
04 · Proof of Truth
Once verified and committed, the data becomes a Proof of Truth certificate. For carbon: a SaveCard that feeds into LCO₂ pre-finance issuance and eventual VCO₂ verified credit conversion. For AI compute: a hardware-verified disclosure record that satisfies SEC climate disclosure, EU CSRD, and SBTi reporting requirements.
Companies use these certificates for radical transparency and regulatory compliance. Registries integrate them into existing methodology workflows. Buyers verify them on chain without trusting any intermediary. The chain of custody from device signature to ledger record is publicly inspectable and cryptographically continuous.
Open data, open assumptions.
Mālama publishes every assumption that goes into the dMRV cycle with confidence levels. For the AI Energy dashboard at aipower.fyi, this means cited data sources with type badges, model assumptions with confidence ratings, full calculation formulas visible, and an open contribution form for researchers. If our numbers are wrong, the system makes it easy to tell us. That is the entire point.
10+ Data Sources
Every calculation references published research, vendor specifications, or peer-reviewed studies. Type badges identify whether each source is academic, vendor-supplied, regulatory, or community-contributed.
6 Model Assumptions
Each assumption is published with a confidence rating. Low confidence assumptions are explicitly flagged. Researchers can challenge any assumption through the contribution form.
Living Dataset
The aipower.fyi dataset is a public good. The contribution form accepts better data, alternative methodologies, and direct corrections. Improvements are versioned and attributed.