How we decide, and what we can't see
This page is the full account of Nero's methodology: scoring rules, evidence handling, validation and the known limits of the data. Product pages summarize; the detail lives here.
How we decide
Nero evaluates what a vessel does. Each track is segmented into episodes, indicators are applied, and every indicator that fires adds points. The total sets the alert level.
Every alert lists the indicators that fired, the points each added and a plain-language rationale. Points and thresholds are set on development data and frozen before independent testing.
Worth monitoring. Keep watching; no action required yet.
Someone should review now and, if needed, contact the vessel or the coast guard.
Behavior strongly consistent with a risk to the asset. Act immediately.
- Fishing is scored separately
Fishing-like loitering is capped at Watch unless trawling speeds appear inside a no-trawl zone.
- Identity anomalies add weight but never trigger alone
Renames, shared numbers and other identity questions can raise a behavioral alert's level, but they never trigger Critical on their own.
- Rules first
In the first phase, every decision comes from fixed, reproducible rules. The same data always produces the same result.
Judged by behavior, never by flag
Flag and nationality carry no weight. An MMSI that contradicts the declared flag is treated as an identity question, never as a nationality score.
Before launch, an automated test will alter a vessel's flag and country code and confirm that its score does not change.
Evidence you can verify
Source data is never altered once written. It is sealed with SHA-256 hashes and recorded in a hash-chained manifest. Every gap in our own reception is logged and never counted as a vessel going dark.
Every report lists its datasets, sources and licenses, file hashes, and code and configuration versions, so the result can be re-run from source data.
Validation before claims
Test data was partitioned and locked before any rule was written. Rules are frozen, then evaluated on held-out data they were never tuned on. Every evaluation run is logged, and misses are reported alongside hits.
We also test the rules on physics-based simulated tracks, and we fix the test cases and the rule variants we will compare before we look at any result.
We will not publish performance figures until official asset coordinates, independent test results and sufficient live data are all in place.
What we can't see
Every data source has blind spots. We list the known limits here and disclose them in every report.
- Vessels without AIS
Vessels that do not transmit AIS are invisible to AIS-based monitoring. Closing that gap requires radar, satellite imagery or other sensors.
- AIS cannot see an anchor
AIS reports position, speed and heading, not where an anchor lies. We therefore say “behavior consistent with a dragging anchor”.
- Anchors dragged while under way
When a vessel drags its anchor while still under way, the change in speed can be slight, and AIS alone often cannot detect it in time. Here we work alongside the cable owner's own sensing, such as fiber monitoring, and Nero identifies the vessel behind the disturbance.
- Reception gaps
Every data source has areas and periods it cannot receive. Each report states the areas we cannot see.
- Self-reported identity
Names, vessel types and identifiers are broadcast by the vessel itself and can be wrong or spoofed.
- No inference of intent
We describe the behavior in the data. We do not judge motive or liability.
Data sources and licenses
Nero does not operate its own sensors. It works with licensed data, partner data and data customers already hold, such as radar or cable alarm systems.
Every record carries source and license tags that determine who a result may reach. Customer data serves only that customer unless the contract says otherwise. If any one source drops, the system keeps running on the rest and reports what it can no longer see.
Where AI fits
Alerts are produced by fixed rules, not by AI. AI assists only with matching vessel-name spellings and drafting report text, and every sentence it drafts must cite evidence and pass automated checks.
For partner data, only vessel names and identifiers are ever sent to external AI services, never positions or tracks.
Privacy and corrections
We do not profile people. Devices that may belong to an individual, such as handheld units or personal rescue beacons, are excluded from vessel dossiers. When we monitor the news, we keep links and our own summaries, not personal names.
If you believe we have a vessel's identity wrong, tell us with the details you have.
We are seeking a small number of cable and offshore wind operators as partners for the first pilot deployments. Inquiries in English or Chinese.