How a report becomes a reliable alert
The full criteria by which every piece of information is gathered, classified, verified and expired. Published so anyone can judge how solid it is.
What this is, and what it is not
Sentry View is a georeferenced record of security incidents, humanitarian situations, political events and infrastructure disruptions, plus a territorial assessment layer that expires and is re-verified.
It is not an early-warning service with guaranteed coverage, and it does not replace any organisation’s own security channels. The absence of a report on the map does not mean nothing happened: it means nothing reached us, or nothing has been logged yet.
No decision that puts a person at risk should rest on this information alone. It is meant to complement the judgement of people who know the ground, not to replace it.
Where the information comes from
Every report states its provenance, and that provenance is shown on the incident card. There are four possible origins:
- Direct observation — material gathered first-hand, including aerial observation. Where there are photographs, the capture date is kept, and the sensor position when the file carries it.
- Cited external source — media outlets, organisations or public reports. The card always links to the original source; third-party content is never copied.
- Information received through direct contact — from people or organisations on the ground. Logged without identifying the sender.
- Automated news intake — a scheduled process reads headlines from a fixed set of outlets and places them in a separate panel, clearly kept apart from verified reports. A headline is not a report: it is a lead to investigate.
Outlets we read headlines from
Haiti: Haïti Libre, Ici Haïti, Juno7, AyiboPost, Haïti Liberté, Le Quotidien 509, Haiti24, and ReliefWeb (OCHA).
African regions: AllAfrica, BBC News Africa, Africanews, RFI Afrique, News24, Mail & Guardian and Premium Times.
Headlines are kept for 60 days and then deleted automatically.
The three confidence levels
This is the core of the method. Every report carries one of these three values, and the criteria are the same for all of them:
| Level | What it means exactly |
|---|---|
| Confirmed | Verified by direct observation, or corroborated by two sources independent of each other — independent meaning neither relies on the other. Two outlets running the same wire copy count as one source. |
| Probable | A single source considered reliable, or consistency with other reports from the same area and time, without independent verification. |
| Unverified | Received and logged, not corroborated. This is the default: a report starts unverified and only moves up when something justifies it. Never the other way round. |
Confidence can go down. If something previously treated as confirmed stops holding up, it is reclassified and the change is recorded in the incident timeline.
Categories
Four categories, each with subtypes so that most real events can be classified precisely instead of falling into “other”:
- Security — clashes, kidnappings, extortion, sexual violence, security operations, arson, among others.
- Incidents — flooding, landslides, earthquakes, storms, power outages, fuel shortages, accidents.
- Humanitarian — displacement, disease outbreaks, malnutrition, child protection, education disruption, blocked humanitarian access.
- Politics — protests, elections, judicial decisions, legislation, armed-group territorial control, international intervention.
Validity: why information expires
This is the main difference from a historical event archive. An area assessed as safe three weeks ago says nothing useful about today, and presenting it as current would be worse than not showing it at all.
So every territorial layer and every route segment carries an expiry. When it is created, a validity period is chosen according to the type of information: from 15 minutes for traffic status to 30 days for a stable territorial assessment. The usual value for a zone is 6 hours.
When a layer expires it does not disappear: it is visually marked as “needs review” and stops counting as verified. It stays visible because an expired assessment may still be the best available data — but you can see clearly that it has expired.
Re-verification
Re-verifying a layer leaves a record with the date, the outcome (no change, improved, worsened, or could not verify), the method used and a note. That history can be consulted from the layer itself. Editing the name or the geometry does not renew verification: that requires an explicit re-verification.
What we do not publish
- Locations of humanitarian organisations’ offices, residences or teams.
- Names of people who report, and anything that could identify them.
- Exact coordinates of reports for anyone without a verified account: the Viewer role receives the position rounded on the server, not in the interface. That person’s browser never receives the real point.
- Attribution of territorial control to groups or individuals, outside moderation accounts.
Corrections
If you find an error — a wrong location, an unfair classification, a misattributed source, a zone that no longer applies — write to gerencia@sentryuav.app naming the report or layer. Corrections are applied to the original record and remain visible in its timeline: the history is not erased.
Known limitations
Stating them is part of the method, because they shape how the map should be read:
- Coverage is uneven. There are more records where there is more connectivity and more sources. An area with few reports may be a quiet area or an uncovered one, and the map alone cannot tell the two apart.
- The team is small. The project is run by one person with occasional support. That limits how much can be verified and how fast.
- Verification is documentary, not forensic. We do not perform image analysis or technical geolocation of received material.
- Expansion into African regions is a pilot. Today it consists mainly of tracking public sources; the verification standard used in Haiti is not yet replicated there, and we do not want to imply otherwise.