Kairos · Threat Math

The Math Behind the Triage

Every tip flows through three mathematical layers — lexical scoring, emotion vector analysis, and a Bayesian Monte Carlo posterior. This page animates each layer in real time using the selected transcript.

Live transcript

Hi I need to report something anonymously . There is a student at Westbrook Academy who has been telling kids he is going to do something serious next week+2.1 . He showed a photo+1.7 of what looked like a weapon+3.0 on his phone to multiple+1.3 people in my class . We are all scared+1.5 . This has been building for the past few weeks and the teachers don't know .

Highlighted tokens carry log-likelihood-ratio weights. Hover-color = category.
1 · Threat-Word TF-IDF Score

Each lexicon word has a pre-computed log-likelihood-ratio (LLR) weight: log P(word | threat) / P(word | benign). We sum a TF-augmented score over every match in the transcript, then squash to [0, 10].

slex=10·(1exp((1 + log·tfi)·wi/ 6))
Worked example for this transcript
weaponweapon+3.00
temporalnext week+2.10
evidencephoto+1.70
evidencemultiple+1.30
distressscared+1.50
s_lex7.98 / 10
Lexicon (live)
gun3.2
knife2.8
weapon3.0
bomb3.6
shoot3.4
kill3.1
attack2.6
fight1.8
hurt1.6
threat2.2
tomorrow2.4
today2.0
next week2.1
planning1.9
photo1.7
video1.7
witness1.4
multiple1.3
scared1.5
afraid1.4
panic1.6
suicide3.0
end it2.7
bully1.6
harass1.7
Math: TF-IDF (Salton 1971) · Cosine similarity (Singhal 2001) · Bayesian inference (Bayes 1763) · Monte Carlo (Metropolis & Ulam 1949).