Back to app index

Active Inference Procurement — Cliff Shed

Procurement commitments as actions that reduce expected free energy: they can be self-fulfilling (stabilize supply, tame variance, make the plan true) or self-deluding (force reality to match a wrong model by spending down optionality, hiding error, and creating unsafe schedule pressure).
Offline / single-file
Press Space to step
Press 1/2/3 for actions

1) World vs Model (and what procurement is “trying to do”)

The model predicts materials + time + risk; procurement acts to make those predictions come true.

Active inference split

Epistemic actions reduce uncertainty (probe supply, test anchors, sample weather window). Instrumental actions push the world toward preferred states (reserve stock, lock prices, pre-stage on the ledge).

Failure mode

Self-delusion happens when commitments mask prediction error (sunk-cost rationalization) or amplify hazard (schedule compression on a cliff). You “win” the forecast but lose the project.

Shed Pre-staged stock (ledge) Supplier Timber need vs stock —
Dissonance (model ↔ sensors) — pushes attention
Self-fulfilling vs self-deluding index (heuristic)

State snapshot

Quantity World Model Error

2) Procurement Console (commitments as interventions)

Choose a procurement move; watch whether it stabilizes reality or forces a hallucination.

Walkthrough

Run a few steps, then try Bulk commit early. If your model is right and the supplier is reliable, this becomes self-fulfilling. If your model is wrong or demand shifts, you’ll see delusion: inventory/cash looks “good” while safety/time degrade.

World knobs (uncertainty + frictions)

Supplier reliability
0.78
Demand volatility (timber needed)
0.35
Observation noise (sensor quality)
0.30
Lead-time (ticks)
5
Cancellation penalty (locks you in)
0.55

Model knobs (priors + stubbornness)

Prior optimism (underestimates need)
0.12
Learning rate (updates from sensors)
0.28
Sunk-cost weight (keeps plan fixed)
0.55

Action queue (commitments in flight)

Order Arrives Qty Lock-in

Event-file log

Capture: each action creates an event-file with context → commitment → outcome.
Explainers (buttons as a guided proof sketch)

1) Probe supply is epistemic: you pay a small cost to sharpen beliefs about lead-time/reliability; it reduces future prediction error without locking you in.
2) Hedge order is mixed: it reserves some stock while preserving options; it lowers variance but can still become delusional under high sunk-cost weight.
3) Bulk commit is instrumental: it pushes the world toward your model by reserving capacity and pre-staging inventory. It becomes self-fulfilling when the model is right and frictions dominate; self-deluding when the model is wrong and commitments hide the error (especially with cancellation penalties and schedule pressure).

3) Mini tutorial: make it self-fulfilling, then make it self-deluding

A pair of controlled “thought experiments” in one screen.

What to watch

Track three coupled quantities: (i) prediction error (dissonance), (ii) option value (lock-in), (iii) safety/time (hazard + schedule pressure). Self-delusion often shows up as “green inventory” with “red risk”.

Why the cliff matters

On a cliff, logistics and safety are tightly coupled, so commitments can alter the feasible set: pre-staging stock reduces hauling exposure (good), but schedule compression under wrong priors increases fall/anchor risk (bad).

Active inference translation for leaders: treat commitments as interventions that reshape the future sensor stream. Good governance distinguishes (a) actions that improve epistemic state (better beliefs) from (b) actions that merely force a narrative to remain coherent.
Save this file as active_inference_procurement.html and open locally; it runs offline.