Argus — Satellite Early Warning
Ground-change detection across Texas. New construction where there was none, at the rural edge of growing cities, scored and delivered as a feed your existing platform already speaks.
What it detects
A change on the ground is a leading indicator, and it shows up in orbit before it shows up in a permit file, a deed record or a tax roll.
Argus reads Sentinel-2 imagery on a schedule, compares each cell against its own history, and reports where the ground has changed. The primary signal is new construction in areas that were previously undeveloped — the pattern that appears at the rural fringe of a metro before any public record catches up with it.
How it works
Imagery. Sentinel-2, operated by the European Space Agency and published openly. No commercial imagery licence sits between you and the evidence, and no vendor's terms of service attach to what you do with a finding.
Detection. Two methods, deliberately. A Clay v1.5 geospatial foundation model produces embeddings for each cell, and a hand-built change detector runs alongside it. Agreement between a learned model and an explicit rule is worth more than either alone, and disagreement is itself a signal worth an analyst's attention.
Scoring. Each candidate is classified CONFIRM or SUSPECT. The
distinction is published rather than collapsed, because a system that reports
only its confident findings is hiding its error rate.
Confounders are reported, not suppressed
Bare earth looks like construction. So does a harvested field, a staging yard, a modular building delivered in a week, and a site cleared after a storm.
Argus evaluates four confounders on every pair it examines — transient,
modular_kit, post_disaster, heavy_tillage — and ships them attached to the
finding. An analyst sees not only that something changed but which innocent
explanations were considered and how each scored.
This matters more than it sounds. For most of this pillar's life the confounders were computed correctly and then went nowhere, because nothing connected the detector to the dashboard. The signal existed and was thrown away. Finding that is the reason we now treat "was it computed?" and "did it reach an analyst?" as two separate questions everywhere in the platform.
Delivery
STIX 2.1 over TAXII 2.1. The feed has been live since May 2026 as Nexus Prism — Satellite Intelligence (Rural-Build Hotspots).
That means it drops into what you already run: OpenCTI, MISP, EclecticIQ, Anomali, Stellar Cyber, ThreatQuotient all read TAXII 2.1 natively. No adapter, no custom integration, no lock-in. If you would rather have rows, the same findings are available over the REST API or as a scheduled workbook.
Who it is for
Non-profits and community organisations tracking development pressure and land-use change at the metro fringe. This pillar was revived for exactly that engagement.
Financial-crime and sanctions teams. Construction is where value gets placed. Unexplained development, on parcels whose ownership is worth resolving, is a lead — and Argus sits in the same platform as the land and property records and the on-chain tracing you would use to run it down.
Government and public-sector planning — unpermitted development, infrastructure demand, and change at a cadence no field survey can match.
What Argus is not
It is not a national product. Sixteen North Texas counties run at full cadence with statewide Texas coverage seeded behind them. Anything outside Texas is a build, not a subscription.
It does not identify people or vehicles. Sentinel-2 resolution is ten metres per pixel. It sees that a building appeared; it cannot see who put it there, and we would not sell you a system that claimed otherwise.
It is not a permit or title check. A finding says the ground changed. Why it changed, whether it was lawful, and who owns it are separate questions the platform can help answer but Argus alone does not.
Area comes from the change polygon, never the scene footprint. Deriving it from the imagery tile boundary overstates it by roughly six thousand times. We mention it because we made that mistake, caught it, and fixed it — and because any vendor quoting you an area figure should be able to say which one they used.