Energy infrastructure · point-in-time

Physical supply disruptions, dated at satellite acquisition.

Daily series for refineries, terminals and chokepoints from five thermal sensors, Sentinel-1 radar and night lights, plus an event log timestamped when the satellite passed. Detection from the sensors, not from headlines.

Request access How detection works

Facilities covered: 60+
Sensors fused: 7
Event timestamp: UTC, at pass

event log · samplefixed at acquisition

Facility	Acquired (UTC)	Signature	First public report
SlavyanskKrasnodar Krai	13 Sep 10:19	SAR +59.7σ	Same dayGUR, Kyiv Post
SaratovRosneft, Volga	11 Sep 00:07	16 hotspots	Hours laterASTRA OSINT
AfipskyKrasnodar Krai	25 Aug 00:28	Fire on process units	Same dayReuters

2026 · Russian refining strike campaignFull record, misses included, in the Detection Audit

Coverage

Where supply is at risk and the press can't see.

More than 60 refineries, export terminals and chokepoints across Russia, the Gulf and the main export routes. Each facility has its own mapped perimeter, capacity, product and destination.

Thermal: VIIRS on Suomi NPP, NOAA-20 and NOAA-21, MODIS on Terra and Aqua. Radar: Sentinel-1. Night lights: VIIRS day/night band.

Refineries and terminalsStruck in 2026ChokepointsA selection of the 60+ facilities covered

One dataset, three layers.

Measurements first, interpretation last. The daily series and the event log are what the sensors saw. The operating state is a separate, labelled model layer, never the headline.

Layer 1 · measurement

Daily series by facility

For every facility and sensor: hotspots and fire radiative power inside the perimeter, the ratio to the facility's own baseline, radar backscatter, night-light radiance, cloud cover at pass time and whether the facility was visible. A zero under clouds is reported as obscured, never as operating.

Layer 2 · measurement

Event log

Each anomaly with the acquisition time, the time QNT42 had it, the sensors involved, its type (strike, accident or incident) and the first public report. Corroboration is added as a new version of the record, never by editing it.

Layer 3 · optional model

Interpreted state

Operating state, capacity offline and episode length, derived from the layers above by a versioned rule set. Delivered as a labelled model output, useful for a dashboard and never mixed with the measurements.

Today in the console. Point-in-time files and an API come with the dataset release, with facilities linked to their operator and listed parent on the same entity keys as QNT42 Positioning.

How a desk reads the event log.

The signal is the physical detection. Corroboration, when it exists, is attached afterwards and is not required for the record.

Detected from the pass itselfRadar structural change and fire on the process units, flagged at the acquisition time, before any report is filed.
Latency, measured per eventSaratov was detected at 00:07 UTC and public confirmation followed hours later. Every event is measured against its first public report, misses included.
Never back-datedThe detection date is the acquisition date. It is not moved to match later reporting, which is what makes the track record auditable.

The event, not the headline.

News wires and aggregators report once an event surfaces publicly. QNT42 measures the event itself and records when it became knowable.

	News and data aggregators	QNT42 Satellite
Sourcewhere it comes from	Public reports, wires, filings	Satellite measurement: radar, thermal, night lights
Timingwhen you learn of it	After it is reported	At acquisition, with latency measured per event
Opaque zonesRussia, Iran, war theatres	Limited by local reporting	Covered wherever the satellites pass
Recordintegrity of the log	Narrative, revised as the story develops	Timestamped at capture, corrections as new versions
Inferencehow far the claim goes	Confident narratives, unquantified	What the sensors measured, no invented probabilities

Built to survive institutional scrutiny.

A detection is not a single hot pixel. Each one goes through the same steps to separate an attack from routine flaring, attribute it to the right facility and fix it in time.

01Several sensorsA strike leaves three signatures: structural change in radar, fire in thermal and a shift in night activity. A detection is corroborated across independent channels.
02Attributed by perimeterA hotspot counts for a facility only inside its mapped perimeter, which filters out neighbouring plants and field burning.
03Against its own historyEach facility is measured against its own baseline. A refinery that never flares and suddenly burns is not a Gulf complex that flares every day.
04Fixed at captureRecorded at the acquisition time and never back-dated, so latency is measurable and the record is auditable.
05Evidence, not inferenceWhat the sensors measured and the capacity, product and direction affected. No disruption probabilities without a validated model behind them.
06Misses publishedEvery detection is checked against the event record and every miss is listed with its cause: clouds, a small fire, no pass.

Access

Talk to research, not sales.

Requests go straight to the research team. We start with the detection record, the method and the facilities that matter to you. One contract covers every QNT42 dataset.

Or write to hi@qnt42.com.