← LibraryAlgorithms for Decision MakingProject Management · Project Risk ManagementLesson 6/18← PrevNext →
ArticlePublished 6 Jul 20266 min readBy Kevin Joginrisk managementdecision theorydecision under uncertaintyBayesian networks
Project Management›Project Risk Management›Series
A risk-focused field guide to deciding well under uncertainty — twenty-six chapters across five parts, re-read for the project professional rather than the roboticist.
The premise
Risk management is decision-making under uncertainty. That is its entire job.
This series works through the modern computational toolkit built for exactly that problem: how a rational actor should choose when outcomes are probabilistic, information is incomplete, and the situation unfolds over time. The mathematics comes from artificial intelligence and operations research — but the reading here is deliberately practical. Less “how to fly an aircraft autonomously,” more “how to decide well when you cannot see the whole board.”
Each chapter explains one method in plain terms, then draws the line back to the work a project professional actually does: quantifying exposure, sequencing responses, learning from outcomes, acting on imperfect status, and coordinating with everyone else who is also making decisions.
How the series is organised
Each part takes away one more thing you are allowed to know for certain, then asks how to decide anyway. Climb it and the problem gets harder — and closer to how real projects actually feel.
Quantify belief and pick the best single choice — probabilistic reasoning and simple decisions.
Decide across a whole timeline, not one moment — Markov decision processes.
Learn the odds while you act — reinforcement learning and exploration.
Decide on belief, not observation — partially observable problems.
Act strategically among many actors — multiagent systems.
The series
Every card names one method, a one-line synopsis, and its risk lens — what the idea means for real project decisions.
Quantify what you believe before you decide.
Encoding beliefs as probability distributions and Bayesian networks — a shared, explicit picture of what drives uncertainty.
Computing the probability of outcomes given the evidence you have observed.
Fitting a model’s numbers from data rather than guessing them.
Discovering which factors actually influence which — learning the shape of the model.
Turning beliefs and preferences into a single best choice, and pricing the value of more information.
Decide well over a whole timeline, not one moment.
Optimal policies when the dynamics are known — Markov decision processes, value and policy iteration.
Scaling to problems too large to solve exactly by approximating the value of each state.
Deciding on the fly by looking ahead from where you are now — forward search and Monte Carlo tree search.
Searching directly for a good decision rule instead of computing values first.
Measuring in which direction a decision rule should change to do better.
Improving a decision rule stably, without lurching or over-correcting.
Pairing a value estimate (the critic) with a decision rule (the actor) so each improves the other.
Testing a decision policy against many scenarios before you trust it in the real world.
Decide well even when you don’t know the odds.
Balancing learning something new against acting on what you already know pays off.
Learning a model of the world from experience, then planning against it.
Learning good actions directly from outcomes, without ever building an explicit model.
Learning a policy by watching an expert rather than by trial and error.
Decide well when you can’t even see the true situation.
Tracking a probability distribution over hidden states as evidence trickles in.
Optimal decisions under partial observation — the partially observable Markov decision process.
Pre-computing a robust policy in advance so you’re ready whatever you observe.
Planning under partial information in real time, from your current belief.
Compact, executable decision policies simple enough to run and audit.
Decide well when others are deciding too.
Game-theoretic reasoning about actors whose choices affect yours.
Decisions taken over time among many interacting actors.
Partial information shared unevenly among many actors, each holding private knowledge.
Aligning independent decision-makers toward a genuinely shared objective.
Where to start
You don’t have to read all twenty-six in order. Pick the route that matches the decision in front of you.
Enough to quantify exposure and sequence decisions well. The backbone every risk practitioner should hold.
For work where the model is unknown or the true state is hidden — the hard, realistic middle of most projects.
The complete ladder, bottom to top, as the field itself builds it. Best read over time, one part per sitting.
About this series
Every explanation, diagram, and risk lens in this series is original KEVOS® work. Nothing is reproduced verbatim. The subject matter is the established field of computational decision-making — probabilistic graphical models, Markov decision processes, reinforcement learning, partially observable problems, and multiagent systems — as consolidated in the standard graduate text by Kochenderfer, Wheeler and Wray. The value KEVOS® adds is the translation: taking a rigorous but abstract body of theory and reading it through the single question a project professional keeps asking — how do I decide well when I cannot see the whole board?
Each chapter page stands on its own, favours intuition over notation, and closes with what the method means in practice. Where theory would help but isn’t essential, it’s noted, not dumped.