Scientific Digital Trust Platform
Every Digital File
Leaves Evidence.
FeXsics reveals the scientific fingerprints hidden inside images, videos, and documents — transforming uncertainty into evidence-backed truth.
42
forensic signals
0.9%
false positive rate
11s
median analysis
The human eye sees an image.Science sees thousands of measurable signals.
Lighting
Illumination consistencyEvery light source casts a measurable direction. We solve the scene's illumination field and test whether each surface obeys the same physics.
Eight disciplines. One conclusion.
The engine runs every discipline in parallel against the same specimen, then reconciles their findings. Disagreement is surfaced, never averaged away.
AI
p 0.04Ensemble detectors trained on generator families, reported with calibrated uncertainty rather than a verdict.
A verdict is not evidence.
Traditional AI detection
- Probably fake
- Black box
- No explanation
- Score without source
FeXsics
- Investigates evidence
- Scientific reasoning
- Transparent findings
- Defensible conclusions
A FeXsics conclusion can be read, challenged and reproduced by another examiner. Every finding carries its measurement, its method and its uncertainty — the standard any court, newsroom or regulator should expect.
Run an investigation.
Load a specimen and watch the examination proceed, signal by signal. This demonstration runs entirely in your browser — nothing is uploaded.
Metadata
Lighting
Compression
Noise
Reflection
AI Analysis
Evidence Graph
Trust Score
Nine stages, one chain of custody.
Upload
Specimen received and hashed
Acquire Evidence
Bit-exact copy, custody recorded
Extract Features
Pixels, frames, containers, streams
Physics Analysis
Light, geometry, optics, materials
Metadata Analysis
EXIF · XMP · C2PA · container
AI Analysis
Calibrated generator detectors
Consensus Engine
Reconciliation and conflict surfacing
Trust Score
Scored with uncertainty bounds
Evidence Report
Signed, reproducible, defensible
One instrument. Eight surfaces.
FEXSICSMethods published. Results reproducible.
Publications
FX-Bench · balanced accuracy
Research timeline
2023
Founding hypothesis: physics before classifiers
2024
First consensus engine · 12 signal families
2025
Open benchmark FX-Bench released
2026
42 signals · deployed in newsrooms and courts
Where a wrong answer is expensive.
We are accepting demo requests.
Tell us what you need to authenticate. You will receive a confirmation by email, and an examiner from our team will follow up with a scheduled walkthrough of the platform.
- Response time
- Within one business day
- Format
- 45-minute live investigation
- Materials
- Bring your own contested media