What the game is
The Supply Chain Game is a Jacobs Industries simulation from Responsive Learning Technologies, the same family as the Littlefield games. I played it at UCD in MEngSc Robotics and Intelligent Manufacturing, module MEEN40790 Supply Chain Design and Analysis. Our team name in the game was disc3, a group of four.
You run a company that makes and sells drums of one product. You change settings (how much to make, when to reorder, where to ship from, how much capacity to buy) and the game scores you on one number, the cash balance at the end. Stock and capacity have no salvage value, so every dollar you spend has to come back before the clock stops.
Phase 1 was one region, one factory and one warehouse. Phase 2 was a five-region network. I finished 16th of 22 in Phase 1 ($6,401,484.36) and 5th of 22 in Phase 2 ($25,404,022.03). This post covers the rules, both phases, the Python model I built, and every mistake I can put a cost on.
Game mechanics
Product and demand. Each drum sells for $1,450. Demand arrives daily at a warehouse. An order that finds an empty warehouse is lost. Partial fulfilment is allowed, so only the unfulfilled part of an order is lost.
The clock. The day counter works like an odometer. Day 50.0 means 50 days have elapsed, and the displayed date is the counter rounded down. Inventory is also recorded at fractional days, when stock hits zero or a delivery arrives. One game day takes 13.846 real minutes, so about 104 game days pass per real day.
Production. A factory makes one batch at a time. A batch costs $1,500 fixed plus $1,000 per drum. Batch time is quantity divided by factory capacity (drums per day). Trucks leave only when the whole batch is done, so a 600 drum batch sends three trucks together.
Reorder logic. Each warehouse has an order point R and an order quantity Q. A production order triggers when warehouse stock plus stock in transit plus work in progress is at or below R. Quantity 0 means no further orders and no order cost. An order point of 0 with a nonzero quantity still triggers when stock reaches zero.
Priority. When two warehouses trigger the same factory, priority 2 is served before 1, and 1 before 0. Equal priority is chosen arbitrarily, so you cannot build a round robin.
Shipping. A truck costs $15,000 per trip, carries up to 200 drums and takes 7 days within a region. Between continent regions it costs $20,000, and $45,000 to Fardo. Mail costs $150, $200 or $400 per unit depending on the lane and takes 1 to 2 days. Outbound cost was $150 per drum in Phase 1.
Money. Holding cost is $100 per unit per year. Cash earns 10% effective annual interest, so the daily rate is 1.10^(1/365) - 1 = 0.02612%.
Capacity. One drum per day of capacity costs $50,000 and goes online after a 90 day delay. It is irreversible. You can place orders that will arrive after the game ends, which sounds useful and is not.
Score. Cash balance at the end. "donothing" is an instructor benchmark team that never changes its settings.
The economics
In Phase 1, with a full 200 drum truck, margin per drum works out like this:
| Item | $ per drum |
|---|---|
| Price | 1,450.00 |
| Production | -1,000.00 |
| Outbound | -150.00 |
| Truck freight (15,000 / 200) | -75.00 |
| Batch fixed cost (1,500 / 200) | -7.50 |
| Margin | 217.50 |
After holding and interest the margin is about $140. A unit of capacity costs $50,000, so it needs about 360 to 370 extra drums sold to pay back. That number sits behind every capacity decision.
Lead time is batch time plus truck time. The reorder point is demand during lead time plus safety stock: R = (daily demand x lead time) + safety stock. A (Q, R) policy fits because production has a fixed cost per batch, so you want fewer, bigger orders and a trigger that fires early enough to cover the wait.
Phase 1: one region
One region (Calopeia), one factory, one warehouse. Defaults were capacity 20, order point 300 and order quantity 150. The default lead time is 150 / 20 = 7.5 days of batch plus 7 days of truck, so 14.5 days. At the peak demand of about 68 drums per day, 300 drums last only about 4.4 days. The warehouse sat empty for roughly 10 days of every cycle.
The late start
My group joined about a real day late, so the default settings ran for 840 game days. In that time demand was 31,310 drums, 14,850 shipped and 16,460 lost. That is 53% of demand gone before we touched anything.
Day 844 changes
We took capacity from 20 to 30 (online about day 934). We raised the order quantity from 150 to 200, a full truck, which dropped freight from $100 to $75 per drum. The order point went up to 5,000, and shipping moved to truck.
Demand
Demand is seasonal. It peaks near days 910 and 1275 at about 68 to 71 drums per day and bottoms out near day 1090 at about 10 per day. I fitted the cycle, and the fit predicted the next 600 days within about 5.5%. My simulation matched the game within about 1% at day 840 and about 5% over days 844 to 944.
Cash path
| Day | Cash ($ thousand) | Note |
|---|---|---|
| 1 | 334 | |
| 840 | 3,303 | before our changes |
| 944 | 3,077 | |
| 1002 | 3,624 | |
| 1058 | 3,450 | 6th place |
| 1111 | 2,580 | |
| 1174 | 2,139 | low point, capacity still coming online and the cash being spent |
| 1215 | 3,180 | |
| 1321 | 5,917 | |
| 1400 | 6,406 | |
| 1436 | 6,142 | |
| 1460 | 6,401 |
Lost demand by period
| Days | Demand | Shipped | Lost |
|---|---|---|---|
| 1 to 840 | 31,310 | 14,850 | 16,460 |
| 841 to 1002 | 9,676 | 3,700 | 5,976 |
| 1003 to 1215 | 4,951 | 4,914 | 57 |
| 1216 to 1321 | 6,928 | 4,623 | 2,305 |
| 1322 to 1460 | 4,227 | 3,043 | 1,184 |
| Whole game | 57,344 | 31,193 | 26,151 (46%) |
The period rows are rounded from the game exports, so they do not add up exactly to the whole-game line.
The capacity test
I tested capacity additions of +3 to +15 units with demand scaled from 0.8 to 1.2. Most additions made after day 844 had negative value, because the 90 day delay ate into the selling window. I added +10. In hindsight +15 would have been about $110k better.
The end game stop
Stock still in the pipe when the game ends is worth nothing, so you stop ordering before the end. The model optimum was around day 1400. Stopping at day 1389 would cost about $77k and at day 1380 about $119k, because the trailing batches never sell.
I planned to stop on Friday 30 October. I missed it. I set quantity to 0 at day 1436.42, about 36 game days late, and 757 drums went unsold. I estimate that cost about $0.7M. The on-time model result was about $7.2M, which would have been roughly 6th to 7th. The best teams finished near $7.9M, and donothing finished at $6,013,562.
I missed it because the clock is slow. At 13.8 real minutes per game day it is easy to misjudge which day you will be on at a given hour. For Phase 2 I set each stop against the on-screen day and staggered them.
Phase 1 result: 16th of 22, $6,401,484.36.
Phase 2: the network
Five regions: Calopeia, Sorange, Tyran, Entworpe and Fardo. Three warehouses: Calopeia (serves Calopeia, Tyran and Entworpe), Sorange (serves Sorange) and Fardo (serves Fardo). Three factories: Calopeia (capacity 70.02), Sorange (built by us, capacity 80, ready day 826) and Fardo (15, later 20). We started on 13 November at game day 730, and the game ended on 20 November at day 1460.
The slow start
At day 762, in an analysis note I wrote on 14 November, we were 11th at $2.40M. The top four were at about $7.2M with nothing built. The note had three forecast scenarios for the final result: $25.5M, $22.7M and $26.6M.
Settings over time
| Day | Setting | Value |
|---|---|---|
| Start | Calopeia lane | quantity 400, order point 1,270, priority 1 |
| Start | Sorange lane | quantity 200, order point 840, priority 1 (quantity later 400) |
| Start | Fardo lane | quantity 200, order point 450, priority 1 |
| 760 to 2,000 | Order points | tuned repeatedly on both main lanes |
| 792 to 827 | Calopeia factory to Sorange warehouse | temporary lane while the Sorange factory was being built |
| 809 | Demand blocking | Fardo and Sorange warehouses blocked from other regions |
| 854 | Demand blocking | Calopeia warehouse blocked from Sorange and Tyran |
| 901.82 | Fardo factory | +5 capacity for $250,000, online day 991 |
| 1001 | Demand blocking | Tyran allowed from Calopeia, blocked from Sorange |
| 1163.85 | Backup lane | Calopeia factory to Sorange warehouse, order point 2,000, quantity 400, priority 0 |
| Late game | Sorange order point | raised to 1,500 |
The model backed the Fardo capacity add at about +$170k.
The backup lane that never ran
I set up the Calopeia to Sorange backup lane at priority 0. It never produced a single batch. The Calopeia factory was saturated by its own priority 1 lane, so priority 0 never got a turn. I had overestimated its value. I should have checked for a produced batch before trusting it.
What the model said
Both main factories ran flat out. The lost demand was structural: it came from capacity, not from policy. Order point variations were worth under $80k, which is inside the noise of the model. Capacity additions after day 1000 were negative. The shipping mode of the last batches made no difference.
Staggered stops
This was the main win. Instead of one stop for the whole network, I stopped each lane at its own optimum.
| Lane | Stop day |
|---|---|
| Calopeia factory, Calopeia row | about 1403 |
| Fardo factory, Fardo row (quantity 0 and order point 0) | about 1414 |
| Calopeia factory Sorange row, and Sorange factory Sorange row | about 1431 |
The last stop landed about one game day later than planned (day 1430). I estimate that cost under $50k. In the model, staggering was worth about $2.9M compared with never stopping. The rule I held to was to trust the on-screen day, never the real clock.
Ranking path
We were 11th at $2.40M on day 762 and first on 17 November. At day 1426 we were 9th at $21.74M, and at day 1431 10th at $22.64M. We finished 5th. In the last 29 days we gained $2.77M, about $95k per day. Alpha gained $2.2M and thesuppliers13 gained $0.63M in the same stretch. At day 1431 alpha was $1.45M ahead of us and thesuppliers13 was $2.85M ahead.
Final top of the table
| Team | Final cash |
|---|---|
| alpha | $26.30M |
| thesuppliers13 | $26.12M |
| 50shadesofsupply | $25.99M |
| supply14 | $25.99M |
| disc3 (us) | $25.40M ($25,404,022.03) |
| spy | $25.32M |
| team15 | about $25.32M |
The model gave $25.1M on its own demand, $26.7M on current demand and $27.5M on high demand. The actual result was $25.40M.
The model
I wrote a day-by-day Monte Carlo simulation in Python. Each run steps through the game one day at a time:
for day in game_days:
demand = sample(demand_model[region], day)
for wh in warehouses:
sold = min(demand[wh], wh.stock) # partial fulfilment
lost += demand[wh] - sold
wh.stock -= sold
for wh in warehouses:
position = wh.stock + wh.in_transit + wh.wip
if position <= wh.R and wh.Q > 0:
factory[wh.source].queue.append(wh)
for f in factories:
if f.idle and f.queue:
wh = max(f.queue, key=priority) # priority 2 before 1 before 0
f.start_batch(wh.Q) # time = Q / capacity
deliver_arrived_trucks(day) # batch done + truck days
cash = cash * (1 + daily_rate) - holding_cost - batch_costs - freightIt tracks stock, in-transit and work in progress per warehouse, a priority queue per factory, batch time, truck lead time, holding cost and interest. The later version had per-warehouse stop and mail_from options. I recalibrated it against each new data export. It runs about 4% conservative. It assumes demand fades over days 1430 to 1460, and that Calopeia seasonal demand falls after about day 1340.
I used it to rank policies, test capacity additions and find stop days.
It could not know the other teams' settings. I only saw their cash, so any story about why they beat us is inference.
What went wrong
Joining a day late cost the most in Phase 1: defaults ran for 840 days and 53% of demand was lost. The late Phase 1 stop cost about $0.7M. Choosing +10 capacity instead of +15 cost about $110k. In Phase 2 the backup lane never ran a batch, the final stop came about one game day late (under $50k), and we started slowly, 11th at day 762.
Why not top 3
The gap to alpha was about $0.9M and to thesuppliers13 about $0.7M. They were ahead through the middle and I recovered fast at the end, gaining $2.77M against their $2.2M and $0.63M, but it was not enough to close the gap.
Structural lost demand and early capacity timing look like the biggest causes. Both are decided in the first few days, and we spent those days at 11th place. I only saw their cash, not their settings, so this is inference, not proof.
Lessons
Capacity is irreversible and takes 90 days to arrive, so decide early. Set stop dates per lane, not for the whole network, and set them against the on-screen day. Do not trust a lane you have not seen produce a batch. Build the model first, then act. Order point tuning inside the noise is not worth chasing.
| Takeaway | Evidence |
|---|---|
| Decide capacity early | +15 would have beaten +10 by about $110k in Phase 1 |
| Stop each lane on its own day | Staggering worth about $2.9M in the model |
| Trust the on-screen day | One missed stop cost about $0.7M in Phase 1 |
| Verify every lane | The priority 0 backup lane never ran |
| Ignore changes inside the noise | Order point variations worth under $80k |
Frequently asked questions
What is the supply chain game?
A Jacobs Industries simulation from Responsive Learning Technologies, in the same family as the Littlefield games. You run a company that makes and sells drums of one product and are scored on the cash balance at the end.
How is the score decided?
By cash at the end of the game. Stock and capacity have no salvage value, so spending that does not pay back before the end lowers the score.
Why stop ordering before the game ends?
Batches that finish after the last sale never turn into cash. In Phase 1 stopping at day 1389 would cost about $77k and day 1380 about $119k.
What is a (Q, R) policy?
A reorder rule. When stock plus stock in transit plus work in progress falls to the order point R, the warehouse orders a fixed quantity Q.