Traditional sandboxes compare business results,
the MulanXu sandbox compares decision behavior.
Multi-agent repeated game + incomplete information. Several groups compete against each other, the market upheaves once every 3 rounds; black swans create real pressure, observing "what reason in good times turns into in people". We don't watch who wins—we watch how you win, and how you lose.
Before comparing business results, look first at decision behavior
One session spans 12 quarters, the market upheaves once every 3 rounds, and black swans fall at random. The result is only Track One; what truly gets recorded is Track Two—how you deform under pressure.
②The game-theory core
Multi-agent repeated game + incomplete information: several groups compete against each other, the market upheaves once every 3 rounds; black swans create real pressure, observing "what reason in good times turns into in people".
Reference: trigger strategies in repeated games—winning once is luck, winning three times is structure.
③Players / Grouping / Duration / Venue
| Item | Specification |
|---|---|
| Grouping | 4–6 players / group · 2–6 groups |
| Players | 12–36 in total |
| Duration | 180 minutes (12 rounds) |
| Venue | Indoor + big screen + host console |
⑦Output
- Track One (revealed first): rounds survived / closing market value / market share / cash / morale → the room cheers or sighs.
- Track Two (revealed after): six-decision-dimension portrait + pressure-state five-dimension change → the room goes quiet.
For the chairman, three charts: the leadership decision-structure radar (all six dimensions overlaid) / the pressure curve (decision-quality fluctuation across 12 rounds) / the vacancy map (which type of decision no one claimed).
Five metrics: one-voice index / idle index / bandwagon index / risk concentration / vacant dimensions.
⑧Consequence mechanic
- Attrition cannot be refilled—a real manpower gap.
- The financing window closes—a real funding constraint.
- Settlement is not judged by the host—the system executes it automatically, to avoid "reasoning it out".
⑨Ending tiers · Contrasting scripts for two groups per session
Mirror
Won, and the structure can be reviewed.
"You won, and you won in round 7—given the same situation again, you'll still win."
Iron Armor
Won, but won on luck.
"You won, but you won on that one gamble in round 7. With this structure, winning once is luck; winning three times is what we call ability."
Unseen
Lost, and with no consensus mechanism.
"You lost, but you're the only group in the room that acted every round only after reaching consensus. Short term that looks slow, long term it's steady. If the cycle stretched to 20 quarters, I'd bet on you."
⑩Three lines of insight
How you deform under pressure matters more than whether you can hold.
Decision structure (who leads, who is absent) is more worth reviewing than the decision's result.
In 3 hours, see in advance how the leadership really looks under market upheaval—this isn't a team-building budget, it's a health checkup for decision risk.
⑪Deployment checklist
Shunxi · 12-Round Decision Sandbox
12 rounds × a five-step decision flow · 4–6 players / group · 12–36 in total · 180 min · 15 minutes per round
④Round flow · five-step decision flow (15 minutes per round)
| Step | Duration | Content |
|---|---|---|
| Intelligence | 2 min | The big screen releases 3 pieces of market intelligence, at least 1 of them vague or self-contradictory |
| Decision | 8 min | In-group discussion, filling in this round's 4–6 decision points |
| Settlement | 2 min | System settlement; the big screen updates all groups' six metrics |
| News | 2 min | Broadcast-style announcement of the quarter's key events and live rankings |
| Record | 1 min | The observer records this round's in-group behavior signals |
Opening six metrics · Black swan cards
Opening (identical for every group, to ensure fairness): Cash 100 / Market share 20% / Team morale 70 / Tech reserve 30 / Brand 50 / Capacity 60.
Black swans (one triggered at random every 3 rounds):
- Sudden policy shift (compliance costs double)
- Loss of a major client (revenue −30%)
- Technological disruption (main product line loses 40% of value)
- A core executive is poached (this group loses headcount and cannot refill)
- Capital winter (the financing window closes)
- Public-opinion crisis (brand −25%)
⑤In-game currency and payout table
| Currency / Metric | Opening baseline |
|---|---|
| Cash | 100 |
| Market share | 20% |
| Team morale | 70 |
| Tech reserve | 30 |
| Brand | 50 |
| Capacity | 60 |
Cash + six metrics: each round's decision points affect the metrics, and the metrics affect the resources available next round. The settlement formula is executed by the host console system, with no manual number changes, ensuring reviewability.
⑥Observation points and signal table
| id | Behavior description | Weight |
|---|---|---|
bet | Goes all-in in a critical round | Courage +2, Wisdom −1 |
conservative | Conservative throughout, never bets | Wisdom +1, Courage −2 |
consensus | Acts every round only after reaching consensus | Connection +2, Responsibility +1 |
onespeech | One person dominates this round's decision (triggers one-voice) | Wisdom +1, Connection −2 |
blame | After a black swan, attributes it to others | Connection −2, Responsibility −2 |
compose | After a black swan, reorganizes the plan within 2 minutes | Growth +3, Responsibility +2 |
silent | After a black swan, says nothing for the whole round | Courage −2, Connection −2 |
longview | Sacrifices current profit for long-term metrics | Wisdom +2, Growth +2 |
Dimensions: full five-dimension coverage + six decision dimensions + pressure-state delta.
HostHost script
- The opening states just one sentence: "This isn't team building, it's a decision checkup." Then collect phones, hand out decision cards, and explain no more of the gameplay.
- Intelligence step: the big screen releases 3 pieces of intelligence, at least 1 vague or self-contradictory; the host reads only, never edits.
- Decision step: enforce only the 8-minute hard stop, no hints, no commenting on in-group disagreement.
- Settlement step: settled automatically by the host console system; the host reports numbers only, never explains or defends.
- News step: announces the quarter's key events and live rankings in broadcast style, without judging how high or low the ranks are.
- Record step: the observer ticks the 8 signals, records the original behavior, and makes no personality comment.
- After a black swan lands: no rescue, no buffer—observe who reorganizes the plan within 2 minutes, who goes silent, who attributes it to others.
- Closing: dual-track settlement—Track One revealed first (the room cheers or sighs), Track Two revealed after (the room goes quiet), then into the debrief script card.
- No-right-or-wrong principle: the host states behavior only, never rates, never judges personality; in-game violations (breaking an alliance, private card swapping, impersonation) are not grounds for elimination but are recorded as behavior signals.
⑦The three output layers
| Report | Delivered to | Content | Boundary |
|---|---|---|---|
| Personal report | The individual | Five dimensions / 16 types / six-decision-dimension radar / blind-spot list / ending tier / insight | Visible to the individual only |
| Team report | Team / HR | Team personality map + complementary network + cooperation decay curve + three metrics | Aggregated results only, no names |
| Leader report | Decision-makers | Three charts + five metrics + structural and risk conclusions | No names, no individual raw five-dimension values |
Track One (revealed first): rounds survived / closing market value / market share / cash / morale. Track Two (revealed after): six-decision-dimension portrait + pressure-state five-dimension change.