Bayesian tools for security and threat analysis.
A detector that is right 90% of the time can still be wrong about most alerts. Work out what an alert is worth, who a leak points to, and which node holds a network together.
Security work weighs weak signals against a low base rate. Most alerts are false even from a good detector. Most leads fit more than one explanation, and a network map rests on ties you are unsure of. These tools carry that uncertainty to the end: the chance an alert is real, how the evidence splits across explanations, and which node matters, with how much confidence.
What do you need to find out?
Which explanation does the evidence favour, and how fragile is that?
A posterior over your explanations, with shared sources counted once and room for the one you did not list. Competing Hypotheses · How do you tell which explanation the evidence favours?
Given this alert, how likely is it real?
The chance an alert is a true positive, from the detector's error rates and the base rate. Diagnostic-Test Calculator · If a test is 90% accurate, why is a positive result often wrong?
Which recipient does a leaked document point to?
A probability over recipients, from the details that survived into the leak. Canary Trap
Which node would break this network apart if removed?
The node, or set of two or three, whose removal splits the network most, across every network that fits. Key Player Removal
Who sits at the centre of this network, and how sure are we?
How often each node ranks first across thousands of networks that fit uncertain ties. Centrality Under Uncertainty
Worked examples (synthetic data)
Instruments
- Diagnostic-Test Calculator
- Competing Hypotheses
- Canary Trap
- Key Player Removal
- Centrality Under Uncertainty
- Capture–Recapture Estimator
- Inter-Rater Reliability Audit
- Collection Planner
- Posting Time Estimator
- Poisson Rate Calculator
- Bayesian Risk Calculator
- Conjunction Calculator
Articles
- How do you tell which explanation the evidence favours?
- If a test is 90% accurate, why is a positive result often wrong?