File size: 1,932 Bytes
8f447ab
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
name: business-sim-env
version: 0.1.0
description: >
  Multi-agent business simulation where an AI CEO manages a software company
  across quarters under adversarial conditions. Tests strategic decision-making
  with delayed consequences across 6 real-world business factors:
  profit potential, risk, team capability, market demand, reputation, and resources.

tasks:
  - id: single_quarter_survival
    difficulty: easy
    max_quarters: 1
    description: End the quarter with positive cash flow. Balance risk vs profit.

  - id: four_quarter_growth
    difficulty: medium
    max_quarters: 4
    description: Grow revenue 30% over 4 quarters. Adapt to market phase shifts.

  - id: adversarial_resilience
    difficulty: hard
    max_quarters: 8
    description: Survive 8 quarters with reputation >= 0.6 despite adversarial shocks.

action_space: CEOAction
observation_space: CompanyObservation

api:
  reset: "POST /reset?task_id={task_id}"
  step:  "POST /step?session_id={session_id}"
  state: "GET  /state?session_id={session_id}"
  tasks: "GET  /tasks"
  grade: "GET  /grade?session_id={session_id}"

reward:
  type: continuous
  range: [-2.0, 2.0]
  partial_progress: true
  description: >
    Shaped reward combining net profit, reputation delta, burnout penalty,
    and bankruptcy punishment. Partial credit awarded at each step.

environment:
  agents:
    - name: MarketAgent
      role: >
        Simulates dynamic market conditions (boom/stable/recession phases)
        using a Markov chain. Affects domain demand and project availability.
    - name: AdversarialAgent
      role: >
        Injects realistic business shocks (client disputes, key dev quits,
        budget audits) to prevent the CEO agent learning brittle strategies.
    - name: DecisionAgent
      role: >
        The LLM-based CEO agent. Makes quarterly decisions via step() API.
        Trained/evaluated by interacting with the environment.