SC5 Twin Run: how to play, from 0 to 100

One game, 6 periods, 3 layers. You run a small 3D-printing factory and decide what the machines, the rules and the robots do. Week 7, 8 and 9 content in one place.

Open SC5 Twin Run

1. Join and set up (2 minutes)

Scan the class QR codeYour lecturer clicks Host and shows a QR code. Scan it. Type a first name or nickname, not your student number. Everyone in the room gets the same scenario, so your results can be compared.

Type your student number inside the gameThe field at the top says "Student number (shown on your screenshot only, never saved or sent)". It stays in your browser. It only appears on your result screen so the marker can match the screenshot to you.

The join screen on a phone: a box asking for a first name or class nickname

2. Play six periods

Periods 1 and 2 (Week 7): pick an intelligence cardDashboard is free but shows a wide demand range. ML forecast costs 6 credits and narrows the range to about plus or minus 3, but it can miss. DL vision costs 9 and predicts defects. GenAI parser costs 4, shows next period's orders early, and is sometimes wrong unless you turn on VALIDATION_RULE. Set the governance switches (approval threshold, safety interlock, features shown, privacy ingest).

Periods 3 and 4 (Week 8): rules, approvals, automationFive rules (R1 to R5) turn the data into one recommendation line. Choose MANUAL (nothing applied), ASSISTED (suggested, you confirm) or AUTO (applied unless it needs a human). Big purchases and supplier exceptions stop and wait for your approval click.

Periods 5 and 6 (Week 9): physical optionsCobot post-processing (more capacity, needs the safety interlock), camera inspection (better with the DL card), AGV transfer (faster, but one charging window costs capacity). Operators are shared between printing and post-processing.

Every period: read, decide, freeze, executeSet your purchases, materials, operators, maintenance, inspection, overtime and outsourcing. Press Generate recommendation, then accept or override each field. Press Freeze twin prediction: the twin records what it expects and your plan locks. Press Execute period: the real seeded week runs. The result screen shows both.

3. A worked example (seed K7Q2, period 1)

The rules panel and the recommendation line in the game

Rule R1 fired: stock is below the lowest possible demand, so it suggests one pack of each material. The confidence of 0.90 comes from the ML card, not from being right. In this run the student bought FDM and SLA packs but not SLS, because they own no SLS printer. The screen records STATUS=OVERRIDE(1). The twin predicted 33 good units and net +98 credits; the real week gave 30 and +82. That gap is what you analyse.

Override is not wrong. Matching the recommendation is not right. The game records what you did. Your reasoning goes in your own document.

4. Screenshot the result screen

The result screen: header, data, recommendation, decision, predicted vs actual, KPI table, flags, reflection question

Take one screenshot per period, right after Execute, so all of these are visible: the header (student number, scenario, seed, period, week, topic), DATA and DATA QUALITY, RECOMMENDATION and DECISION with ACCEPT or OVERRIDE, PREDICTED and ACTUAL with the EVENT code, the KPI table (cost, lead time, service, carbon, risk: before, predicted, actual, delta), the FLAGS line and the REFLECT question.

Laptop: Win+Shift+S or Cmd+Shift+4. Phone: power + volume down. Name the files P1.png to P6.png. After period 4 and after period 6 a Resume code appears under the card. Photograph it or copy it by hand: it rebuilds your run in the Week 9 class, on any device. The reflection question has no answer box on purpose. Answer it in your own words.

5. Use it in A2 and A3

Swipe the table sideways to see all four columns.

Assessment step	What you did in the game	Evidence on the screenshot	What you write yourself
A2 step 5: KPIs to analytics, ML, DL	Chose Dashboard, ML forecast or DL vision in periods 1 and 2	DATA line, FORECAST_ERR, KPI before vs actual	Why that card fits your A1 KPI and what its error cost
A2 step 6: GenAI workflow	Tried GENAI_ORDER_PARSER with validation on or off	SIGNAL=OK or CORRUPT, ETH-02 flag	Where a GenAI step needs a validation rule and a human check
A2 step 7: workflow automation and exceptions	Set rules R1 to R5, the threshold, MANUAL / ASSISTED / AUTO, approved escalations	RECOMMENDATION line, STATUS=ACCEPT or OVERRIDE, ESCALATIONS, GOV-01	Your rule design: when automation acts alone and when it must stop
A2 step 8: physical automation	Printers, cobot, camera inspection, AGV in periods 5 and 6	Physical fields in DECISION, lead_time and carbon deltas, AGV_CHARGE event	Which technology you would buy and why, with the numbers
A3 steps 3 to 7: diagram, detect and respond, before and after, numbers, risk strategy	Play the same seed twice: automation OFF, then ON	Two result screens with the same SEED and different KPI rows; EVENT and RULE codes	The SCDT diagram (data, intelligence, rule, approval, action), the before and after comparison, your mitigation strategy

Deadlines: A2 Sunday 20 September (end of Week 8). A3 Sunday 25 October. Play in the Week 8 class; replay periods 5 and 6 with your resume code in Week 9.

About SC5 Twin Run

The concept

A digital twin is a model that runs on the same data as the real operation, so a decision can be tested on the twin before it is executed. Supply Chain 5.0 adds the human: approvals, overrides and judgement sit inside the loop, not outside it.

How it works

Six periods use seeded operating data and fixed rules. In each period you accept or overrule the recommendation; the game then executes physically and compares your run with the frozen twin run on cost, lead time, service, carbon and risk.

Winning strategy

Overrule only when you can name the information the twin does not have. Track which overrides improved the comparison and which just felt safer; the debrief questions at the end ask exactly that.

Core idea: Digital twin