Most dMRV procurement goes wrong the same way: the evaluation focuses on what the software shows, and verification later tests what the software can prove. These eight criteria are ordered by how much they reveal, and every one of them is answerable by a vendor who has nothing to hide.
Ask About Method Before Features
A demo shows outputs. Verification interrogates derivation. The gap between those two is where procurement decisions go wrong, because the parts of a system that determine whether credits issue, namely sampling design and uncertainty propagation, have no visual representation at all.
The eight criteria, in order of what they reveal
Sampling design
How is plot count derived, and from what measured variance? A good answer references a target precision. A bad one gives a number with no derivation.
Uncertainty method
Is uncertainty propagated through every conversion, or attached at the end? Ask which error term dominates on a real project.
Data traceability
Can a headline tonne be walked back to an individual field record, with change history intact?
Tier capability
Does the system support site-specific measurement, or does it run on default factors behind a good interface?
Ecosystem coverage
Mangrove, peat and dryland need different protocols. A single upland forest workflow applied everywhere is a red flag.
Registry outputs
Which standards has output been accepted under, verified by which bodies? Ask for the verification, not the logo.
Raw data export
Can a third party inspect the underlying records without the vendor mediating?
Local field capability
Who actually runs the plots, and can the protocol be executed consistently by the people who will run it?
Three Architecture Patterns
Platforms tend to cluster into three shapes, each with a characteristic weakness.
| Pattern | Strength | Characteristic weakness |
|---|---|---|
| Satellite-first | Wide coverage, fast change detection | Thin ground sample; calibration range often unstated |
| Field-first | Defensible measurement | Slow, expensive, weak between inventories |
| Hybrid | Coverage plus calibration | Integration quality varies enormously |
The category matters less than whether the ground sample is statistically designed. A hybrid system with a convenience-located field sample carries the same weakness as a satellite-only one, with more moving parts.
Answers That Should Worry You
A proprietary accuracy percentage with no published method. Accuracy is a property of a method applied to a population — a bare number is a marketing claim.
Reluctance to share sampling design, framed as commercial confidentiality. Verifiers will require it regardless; a vendor unwilling to discuss it at procurement is describing a future problem.
Model fit statistics from calibration data presented as predictive accuracy.
And silence on ecosystem differences — a vendor who does not volunteer that mangrove requires different sampling than dryland has not thought about it.
What to Require in Writing
Two commitments are worth contractual weight: raw field data remains exportable and inspectable by you and by any verifier you appoint; and the sampling design, including its derivation, is documented and provided.
Those two clauses cost a good vendor nothing and eliminate the failure mode that actually destroys projects — arriving at verification with numbers nobody can defend.
Image credit
Hero image: Rainforest canopy layer top view by Chamberlain of Nilai, CC BY-SA 4.0, via Wikimedia Commons.
Move beyond estimates
Verifiers test sampling design, uncertainty and whether a number traces back to the field. TREEO dMRV captures that evidence in real time, in one auditable chain.
