From a headline to evidence you can publish

Verify news events with a source-linked location, a full evidence thread, and satellite imagery that can confirm — or honestly fail to confirm — what was reported. Built for journalists, fact-checkers and investigative researchers.

Open the Watchfloor Read the methodology

Quick answer — how Delta supports news verification

Off-Nadir Delta turns a reported event into checkable evidence: near-duplicate articles are collapsed into one event with an independent-source count, the location is AI-corrected and checked against the source text, every source article stays linked, each claim keeps an evidence class over time, and a satellite observability verdict says what open Sentinel imagery could confirm. When evidence is missing, the product says so rather than filling the gap.

The verification loop

1. Find the event behind the headline

Delta Signals collapses near-duplicate reporting into one event with a title, a category, and a count of the distinct outlets carrying it — so "nine articles" and "nine outlets" are never confused. The outlet count is not an independence count: ten outlets running the same wire story count as ten, so independence is reported as not established rather than asserted.

2. Check the location, not just the country

Raw feeds often place an event at country level or at the city it was reported from. Delta resolves precise coordinates, checks them against the source text before mapping, and says openly when a location could only be resolved to an area — an area is shown as an area, not a false point.

3. Read the evidence thread

Every event carries its full thread: when it was first reported, how the story changed, and every source article behind it — each one a link you can open. Attribution (who did it) is tracked as a separate axis from occurrence (did it happen), because nine outlets reporting an event can still rest on one party’s statement for who caused it.

4. Ask what a satellite can settle

Each event carries a satellite observability verdict and a recommended sensor. Sentinel-1 radar works through cloud and at night; Sentinel-2 optical confirms visually. When open imagery cannot resolve the question, the product says so instead of offering an image that could not show it.

5. Audit what you were told

The claim ledger keeps every claim with its evidence class (confirmed / reported / party claim / disputed / assessment) and records when a later answer restated a claim with weaker evidence — the downgrades are exactly the ones to re-check before publication.

For situational awareness and published verification — not for tracking or targeting individuals. All imagery is openly licensed Sentinel-class data you can republish with attribution.

Frequently Asked Questions

How do I verify a news event with satellite imagery?▼

First fix the location and date: an image can only confirm something at a known place and time. On Off-Nadir Delta the event signal carries an AI-corrected, source-checked coordinate and a reporting timeline. Then check observability — whether the event leaves a mark that Sentinel-class imagery (~10 m) can resolve — and compare a pre-event and post-event scene. Radar (Sentinel-1) works through cloud and darkness; optical (Sentinel-2) confirms visually in clear daylight.

What can free satellite imagery actually confirm?▼

Open Sentinel-class imagery resolves regional-scale change: flooding extent, fire scars, large structural damage, vessel presence, and construction over time. It generally cannot identify individual vehicles or people — that needs sub-meter commercial imagery. Off-Nadir Delta states per event what open imagery could and could not confirm, so a verification claim never outruns the sensor behind it.

How is this different from just reading the news feed?▼

Three ways. Near-duplicate articles are collapsed into one event, with an explicit count of the distinct outlets carrying it rather than raw article counts — stated as outlets, because syndicated reprints are not collapsed and so independence is not established by that number. The location is resolved and checked rather than taken from the dateline. And each claim keeps an evidence class over time, so you can see when "confirmed" quietly became "reported" — or the reverse.

Can I cite and publish what I find?▼

Sources are the original news articles, which you cite as such. Sentinel imagery is openly licensed and can be published with attribution ("Contains modified Copernicus Sentinel data"). Every satellite field shown on an event page is the same value the public API serves, so a methods section can reproduce the lookup.

Does the AI make up details when evidence is thin?▼

The product is built to refuse rather than fabricate: when evidence is missing it says what is missing instead of filling the gap. You can see a real refusal, produced by the live system, on the methodology page.