Hive Diagnosis 2D: Bayesian Belief Map

1 · Run a Diagnostic Test

2 · Implement a Diagnosis

Belief-propagation speed 1.0×

Live Diagnostics

Posterior entropy

2.32 bits

Diagnostic cost

Leading hypothesis

— 20%

New case started. No tests run yet — all five causes equally likely.

How it works

This is the flat 2D companion to the 3D belief-network simulator: the same naive-Bayes diagnostic model, redrawn as a top-down belief map instead of an orbiting 3D scene. Each hive "case" has one true underlying cause, hidden from you. Five candidate causes start with a flat prior P(c) = 1/5. Every diagnostic test you run reports a real (simulated) positive/negative result, and the belief over causes is updated with Bayes' rule, treating tests as conditionally independent given the cause:

P(cause | evidence) ∝ P(evidence | cause) × P(cause)
posterior(c) = prior(c) × L(c)  /  Σ_c' [ prior(c') × L(c') ]
L(c) = P(test positive | c)   if the test came back positive
L(c) = 1 - P(test positive | c)   if it came back negative

Uncertainty is tracked with Shannon entropy over the posterior — it starts at log₂(5) ≈ 2.32 bits (maximum confusion) and falls toward 0 as evidence narrows things down:

H = -Σ_c p(c) · log₂ p(c)

This mirrors the article's five-step troubleshooting process: identification (a new case appears with a hidden cause), critical thinking (choosing which test actually reduces entropy instead of guessing), solution development (picking a diagnosis from the dropdown), implementation (the "Implement Fix" button), and evaluation (the reveal — was the fix actually right, and how much did it cost in tests?). Rushing to implement before entropy is low is exactly the "skip identification" shortcut the article warns against.

Cause disks grow and brighten as their posterior probability rises; each test node lights green (positive) or red (negative) once run, and glowing dots carry the belief update from the test to every cause.
Drag to pan the belief map, scroll or pinch to zoom.
Some tests are cheap and broad, others narrow but costlier — the rational strategy is the one that cuts entropy fastest per unit cost, not the one you happen to try first.

Colony hive (case center) Candidate cause (size = probability) Diagnostic test node

Drag to pan · scroll to zoom