1 · Read the stateCold-room stock and the truck arriving today define the decision state.
2 · Order aheadChoose 0, 3 or 6 batches. Today’s order arrives tomorrow.
3 · Learn from samplesTrain ADP on selected trajectories instead of enumerating every state.
4 · Compare fairlyExact DP, ADP and a myopic rule face identical demand streams.
Melbourne cold-chain flow
Every moving batch is tied to the current simulated transition.
1 / 12Decision day
6Cold-room stock
$0Your cumulative cost
100%Demand served
Transition log
- Ready. Choose an order quantity for tomorrow.
Why this action?
ADP has not seen much yet. Train it, or play and let sampled returns build experience.
Exact policy table · order quantity by day and stock
Order 0Order 3Order 6
ADP policy gap
Approximation coverage
ADP groups nearby stock states, then estimates their future value from experience. This trades precision for speed and scalability.
200 shared demand paths
Decision briefing
About this game
The concept
Dynamic programming decomposes a sequence of linked decisions. The value of today’s choice includes its immediate cost and the best future cost from the next state.
DP versus ADP
Exact DP enumerates the state space. ADP estimates value from sampled experience, making larger real-world problems tractable while accepting approximation error.
Supply-chain insight
Replenishment is not a one-shot order. Lead time links today’s action to tomorrow’s service, so a locally cheap choice can be globally expensive.