Decision Security

Reusable decision-science utilities for security — Monte Carlo risk bands, Bayesian updates & calibration, survival helpers, Value of Information, light causal helpers, and visualization.

Part of Apropos Security · Notebooks · Playground · Blog

Install

pip install --pre decision-security

Quickstart

import numpy as np
from decision_security.montecarlo import risk_bands, var_es, make_lognormal_severity, simulate_aggregate_losses

sev = make_lognormal_severity(meanlog=8.0, sdlog=1.2)
losses = simulate_aggregate_losses(n_periods=10000, lam=0.6, severity_sampler=sev)
print(risk_bands(losses))      # {'p50': ..., 'p90': ..., 'p95': ...}
print(var_es(losses))          # (VaR95, ES95)

Modules

• synth — synthetic data (heavy-tail losses, counts, mixtures, survival with censoring, categorical/Dirichlet)
• montecarlo — Poisson frequency + severity, risk bands, VaR/ES
• bayes — Beta-Binomial & Normal(known σ) updates, calibration helpers
• survival — simple Kaplan-Meier & Nelson-Aalen estimates
• voi — Expected Value of Perfect Information (EVPI) and simple ROI selection
• causal — tiny DAG utilities (parents, descendants, naive backdoor set)
• viz — small matplotlib helpers (loss distribution, risk bands, KM curves)

Status

0.x pre-release (APIs may change).

Docs & examples

• Notebooks: Security Decision Science — 19 interactive notebooks using this library
• Playground: Security Decision Labs — FAIR Simulator app
• Hub: apropos-security.com

Contributing

Issues and PRs welcome. For non-public questions, contact via LinkedIn.