GroundTruth AI Decision intelligence from Earth observation
Turn Earth observation into decisions you can defend.
GroundTruth AI combines optical, SAR, terrain, thermal and time-series evidence to rank where field teams should investigate first—and shows the uncertainty behind each priority.
Current statusTRL 4 · Controlled-environment validation complete; external pilot is the next gate.
01 / The decision gap
Imagery is abundant.
Decision confidence is rare.
Teams do not need another unexplained heatmap. They need a short list of investigation areas, the evidence behind each priority and a clear account of uncertainty and data gaps.
02 / The platform
A disciplined path from signal to field plan.
GroundTruth AI helps expert teams turn dispersed geospatial evidence into a ranked field plan. It narrows the search; field verification makes the decision.
01—05
Optical, radar, terrain and contextual evidence—reviewed together.
01Assemble
Build the evidence base
Harmonize optical time series, SAR backscatter, terrain, thermal and approved partner layers.
02Test
Stress-test generalization
Use spatial holdouts, coordinate controls, hard negatives and regional checks to expose brittle performance.
03Rank
Rank with context
Prioritize investigation areas using confidence, environmental fit and traceable contributing signals.
04Act
Fit field workflows
Export GIS-ready layers, evidence tables, limitations and concise reports for expert review.
02.5 / Inside the platform
From regional signal to a reviewable field priority.
The workflow brings ranked areas, contributing evidence, uncertainty and provenance into one review path. The visualization shows multi-satellite Earth coverage; the protected demo uses non-sensitive data.
- 01Define the missionConfirm the study area, model version and evidence coverage.
- 02Review ranked areasMove from a regional mesh to a short list of candidate cells.
- 03Keep the decision recordTrace contributing signals, uncertainty, limitations and provenance.
LANDSAT + SENTINEL / GLOBAL COVERAGEEARTH OBSERVATION
Visualization · NASA SVS ↗
▶
Open the private demoServer-protected access · approved reviewers
03 / Validation, not theatre
The score matters. The test regime matters more.
We publish each result with its validation setting—including promising transfer results and a negative holdout result that changed both our claims and next test.
Performance under a spatially separated audit regime.
Evidence of transfer under a held-out country setting.
A useful negative result—not hidden, and not reframed as field accuracy.
Directional evidence only; the sample is too small for a broad claim.
What changed in the validation program
- 01
Random splits were not accepted as final evidence.
Evaluation moved toward spatial separation and country-out transfer.
- 02
Reproducibility became a product requirement.
Frozen artifacts, score reproduction and file integrity are tracked.
- 03
Failure now shapes the next experiment.
The general-model holdout result narrowed claims and strengthened the pilot protocol.
Important
These results demonstrate validation maturity within documented regimes. They do not establish commercial accuracy in a new geography. Independent, partner-led field validation is the next proof point.
03.5 / Regional-expert validation update
Nine regions. Three independent runs each. One preserved evidence trail.
The July 2026 evidence brief documents a repeatable regional-expert training stage. These internal ranges are reported separately from the spatial-audit and country-out regimes above.
Open the Proof of Evidence brief ↗Evidence boundaryLeakage controls passed in the internal record, but this is not yet an independent blind external test. Partner-held unseen outcomes remain the next proof.
04 / Mission architecture
One evidence core. Five focused mission areas.
02
Environmental monitoring
Prioritize review of landscape change, heat, moisture and risk-oriented signals.
03
Land & engineering
Direct scarce ground-investigation capacity toward evidence-supported areas.
04
Infrastructure resilience
Support structured review across distributed assets and complex operating environments.
05
Cultural heritage
Support expert-led survey prioritization while protecting sensitive location data.
05 / Where the project stands
Internal evidence is documented. External field proof comes next.
Technical foundation
Multi-sensor workflow, reproducible outputs and protected demonstration environment.
Controlled validation
Spatial audit, country-out testing, integrity checks and documented limitations.
Partner pilot design
One bounded, non-sensitive area with a frozen protocol and independent review.
Toward TRL 5
Demonstration in a relevant environment, measured against an agreed baseline.
Partner pilot protocol
A result designed to survive scrutiny.
- 01Partner defines area and baseline
- 02Model, thresholds and protocol freeze
- 03Predictions lock before ground truth
- 04Independent review measures operational value
06 / Updated public materials
Review the latest evidence, progress and partnership path.
All public documents below reflect the current 2026 project position and exclude sensitive locations, credentials and proprietary implementation details.
01PDF · JUL 2026
Public validation & progress
Current evidence, limitations, reproducibility and the TRL path.
Open report ↗02PDF · 2026
Sanitized evidence summary
A reviewable evidence package without sensitive operational details.
Open report ↗03PDF · 2026
Non-confidential partnership deck
Mission, pilot structure, value proposition and collaboration routes.
Open report ↗04PDF · JUL 2026
Proof of Evidence 2026
Regional-expert architecture, controlled results, evidence boundaries and the next blind pilot.
Open report ↗07 / Secure environment
Review the workflow in a protected environment.
The private mission console is available to approved partners, investors and technical reviewers. Access is protected on the server and sensitive code is not exposed.