It recognises recurrence
SIFRA notices when the same pattern of conditions keeps happening — for example, interference and vibration rising together for two minutes, several times a day.
SIFRA records the real conditions around your site — GPS quality, vibration, sound, weather and passing trains, with radio interference ready to add — every second of every day, along with what every sensor saw. When a sensor sees something, you can always check what was happening around it at that exact moment. GHOST HUNTER, URIEL and EIDOLON are all built from this record.
The question it exists to answer
What was happening in the physical environment at exactly the moment a sensor made an observation?
What SIFRA finds in the record
SIFRA notices when the same pattern of conditions keeps happening — for example, interference and vibration rising together for two minutes, several times a day.
SIFRA doesn’t guess what the pattern is. A person names it, for example ‘passing train’, and the record keeps who named it and why.
At 04:12, four different kinds of instrument — radio interference, vibration, GPS signal and sound — all changed at the same moment. That is the finding. One instrument jumping could be a glitch; four unrelated instruments reacting together is very hard to explain away. SIFRA never blends them into a single score. Each reading is kept as it was measured, so anyone can check exactly what each instrument saw.
Naming a moment ‘passing train’ never hides it. Those are often exactly the moments a tester most wants to see.
SIFRA suggests real situations from the record that are worth testing against, such as ‘heavy rain at night’. A person approves each one before it is used.
SIFRA will never tell you it was a train. A person tells it that, and the record keeps their name.
When a drone is flown to test a sensor, SIFRA uses the drone’s own flight log to see where it really was, then checks how well each sensor did.
Results come as a signed file that anyone can verify for free.
Data exports in the formats engineers already use.
You are alerted if an instrument goes quiet.
Buyers can compare different suppliers’ sensors in exactly the same recorded conditions.
Real recordings are replayed to test sensors and software, and to train operators.
Two offers
Install SIFRA at your own site and build your own record. It’s yours. Keep it running, or use it for about six weeks to check a site before you install anything, then move it to the next one.
A licence enquiry, not a checkout. This is a conversation, not a purchase.
Licence enquirySubscribe to the record from our own test site beside a busy electrified railway. It grows every month, and you get all of it.
The record grows every month, and subscribers get all of it.
Someone licensing in the first month receives everything recorded from that point onward as well as everything recorded before they arrived. The archive a subscriber holds gets larger and more useful for as long as they hold the licence. Later subscribers start with more; earlier subscribers have had it for longer.
GHOST HUNTER and URIEL are built from SIFRA’s record of your site, so they always come with SIFRA. GHOST HUNTER spots alerts that look like known false signals. URIEL explains, in plain words, what the conditions are doing to each sensor. They only explain what is actually in the record; they never make anything up.
Artefact recognition. It compares each detection with the false signals SIFRA has recorded at that site, such as multipath, rail traction and interference, and says which it resembles. It ranks and explains; it never hides a detection.
For operators at airports, ports and critical sites. Live condition briefings, the WHY? button, and GHOST HUNTER’s judgements in plain language.
For makers of command systems and sensors. The same insight as data, attached to each detection inside their own system.
URIEL runs at the customer’s own site, on the customer’s own SIFRA record. GHOST HUNTER is part of both editions. Nothing is sent to us.
Condition briefings are available from the first day of recording; GHOST HUNTER’s judgements sharpen as the record grows.
SIFRA continuously records the physical environment in which sensors operate (GNSS, vibration, sound, weather and rail traffic, and ready to integrate radio and electromagnetic interference) and every sensor observation alongside it. It is the record GHOST HUNTER, URIEL and EIDOLON are built from.
The question it exists to answer
What was happening in the physical environment at exactly the moment a sensor made an observation?
What SIFRA finds in the record
When electromagnetic interference and vibration rise together, in the same shape, for the same duration, repeatedly, that is a recurring environmental signature. SIFRA identifies it as one rather than leaving a person to notice.
Its output is an unnamed recurring signature. A name arrives from exactly one of two places, and the record stores which: a named human classified it, with a stated reason — or an external record such as a timetable was matched against it, and the source is named.
Four independent domains departing from their baselines together at 04:12 is the finding, and the independence is its entire evidential weight. Four instruments measuring different physics and agreeing is a completely different claim from one number being large. Each domain keeps its own magnitude, in its own units, with its own observations.
Naming a window does not establish that nothing else happened inside it. To somebody stress-testing a receiver, a window labelled as a rail traction pass is the most valuable window in the archive, not the one to discard.
SIFRA generates candidate scenarios — selections of historical conditions worth replaying — containing no measurement that is not traceable to a stored observation. Whether a scenario is replayable is recomputed every time it is asked, never stored as a flag, and when it is not ready the reason names itself: a missing observation, a recorded gap over the window, a lapsed calibration certificate. A candidate becomes a scenario only when a named human accepts it.
A hundred occurrences of the same pattern establish that something recurs. They say nothing about what it is. SIFRA will not tell you it was a train. A person tells it that, and the record keeps their name.
When a drone is flown to test a sensor, its own flight log shows where it really was. SIFRA imports PX4, ArduPilot, MAVLink and GPX logs directly, and encrypted DJI logs through a declared third-party export. It aligns the log in time, then measures what each sensor under test reported against it, together with the conditions at that moment. Without a log, no detection rate is computed.
A trial’s results leave as a signed pack that anyone can verify with a free tool. A sensor is either a site instrument or a system under test, never both, so no system marks its own work.
The canonical record, plus spreadsheet and data-science tables, radio recordings in the SigMF standard, audio, original video, and robotics logs for replay tools. Every table keeps the provenance of every value, so a table taken out of the archive still says where each number came from.
Sensors using the SAPIENT standard are recorded as site instruments like any other. An alert is raised when an instrument goes quiet, so unattended sites are watched.
Bidders’ sensors can be compared in the same recorded conditions, so a procurement body judges evidence rather than claims. The evidence pack can be verified even without SIFRA.
Recorded windows are replayed to test sensors and software updates against real conditions before release, and to train operators on real events rather than invented ones.
Two offers
Deploy SIFRA at your own site and build your own record, which you own. Installed by your own engineers or an accredited integrator. Run it continuously, or deploy it for around six weeks to characterise a site before sensors or systems are installed, then move it to the next site.
A licence enquiry, not a checkout. This is a conversation, not a purchase.
Licence enquiryA subscription to the environmental record from the EÒLDÌON STRESS TEST ENVIRONMENT: a working industrial site in Gdańsk Letnica, beside an active electrified railway, where GNSS degradation and electromagnetic interference are expected.
The record grows every month, and subscribers get all of it.
Someone licensing in the first month receives everything recorded from that point onward as well as everything recorded before they arrived. The archive a subscriber holds gets larger and more useful for as long as they hold the licence. Later subscribers start with more; earlier subscribers have had it for longer.
GHOST HUNTER and URIEL come from SIFRA’s record of a site. False signals and train passes look different at every site, so each learns from its own site’s record: without SIFRA, neither exists, and URIEL is always supplied with SIFRA. URIEL explains, in ordinary language, what the environment is doing and how it affects each sensor. It invents nothing: it explains only what is in the record.
Artefact recognition. It compares each detection with the false signals SIFRA has recorded at that site, such as multipath, rail traction and interference, and says which it resembles. It ranks and explains; it never hides a detection.
For operators at airports, ports and critical sites. Live condition briefings, the WHY? button, and GHOST HUNTER’s judgements in plain language.
For makers of command systems and sensors. The same insight as data, attached to each detection inside their own system.
URIEL runs at the customer’s own site, on the customer’s own SIFRA record. GHOST HUNTER is part of both editions. Nothing is sent to us.
Condition briefings are available from the first day of recording; GHOST HUNTER’s judgements sharpen as the record grows.