Here is the uncomfortable truth Part IV insists on: you cannot see the true state of your project. You see reports, metrics, and test results — shadows of the state, arriving late and distorted. So you reason about a belief instead.
Parts I–III let you at least observe which situation you were in. Real projects deny even that. The schedule status you're shown is a lagging summary; the "green" on the dashboard may be optimism; the true health of the work is hidden behind imperfect signals. The right object to reason about is therefore not the state but a belief: a probability distribution over which state the project might actually be in, given everything you've observed so far.
1Keeping the belief current: filtering
A belief must be updated as evidence arrives, and the update is pure Bayes. Start with your current belief; account for the effect of the action you took (the project moves on); then fold in the new observation, upweighting the states that would most likely have produced what you saw. The result is your revised belief. Repeated over time, this is Bayesian filtering — and the belief it maintains is a sufficient statistic of the entire history: carry it, and you can throw the raw log away.
2Representing the belief in practice
- For a handful of discrete states, keep the belief as an exact table of probabilities and update it directly.
- For continuous quantities with roughly linear dynamics and noise, the Kalman filter keeps the belief as a Gaussian — a mean and a spread — updated in closed form. Its extended and unscented variants handle mild non-linearity.
- For messy, non-linear, multi-modal situations, the particle filter represents the belief with a swarm of sampled hypotheses, reweighting and resampling them as evidence lands — flexible enough for almost anything.
When you can't observe the true state, the honest object to hold — and to act on — is a belief: a probability distribution over where you might be, kept current by folding each new observation into the last. The belief, not the latest report, summarises everything you know.
Stop treating the status report as the truth and start treating it as evidence about a hidden truth. Maintain an explicit, probability-weighted picture of where the project really stands — and, crucially, of how confident that picture is — updating it as each indicator arrives rather than lurching with every fresh number. A belief that carries its own uncertainty tells you not just your best guess at project health, but how much you should trust it, which is exactly what a good decision needs.