Environment Reference

This section provides detailed documentation for all 20 environments in Coopetition-Gym.

Not sure which environment to use? Try the Environment Finder - an interactive tool that matches your research questions to the most relevant environments.

Action Space Model

All Coopetition-Gym v1.x environments implement the uniaxial treatment of coopetition, where agents choose cooperation levels along a [0, endowment] continuum. This design follows the social dilemma tradition in coopetition research (Bengtsson & Kock, 2000; Lado et al., 1997).

Key characteristics:

Coopetition-Gym v2.x will introduce biaxial treatment with independent cooperation and competition dimensions, enabling agents to simultaneously vary investment in value creation and value capture. See Scope and Strategic Roadmap for theoretical rationale and extension plans.

Environment Overview

TR-1: Interdependence and Complementarity (5 Environments)

Environment Agents Category Key Challenge
PartnerHoldUp-v0 2 Dyadic Power dynamics and exploitation
PlatformEcosystem-v0 1+N Ecosystem Ecosystem health management
DynamicPartnerSelection-v0 N Ecosystem Social learning and signaling
SynergySearch-v0 2 Benchmark Exploration vs. exploitation
RenaultNissan-v0 2 Case Study Multi-phase alliance dynamics

TR-2: Trust and Reputation Dynamics (5 Environments)

Environment Agents Category Key Challenge
TrustDilemma-v0 2 Dyadic Long-horizon impulse control
RecoveryRace-v0 2 Benchmark Planning under trust constraints
SLCD-v0 2 Case Study Validated Samsung-Sony model
CooperativeNegotiation-v0 2 Extended Commitment and breach penalties
ReputationMarket-v0 N Extended Reputation as strategic asset

TR-3: Collective Action and Loyalty (5 Environments)

Environment Agents Category Key Challenge
TeamProduction-v0 4 Collective Action Free-rider dynamics
LoyaltyTeam-v0 4 Collective Action Loyalty-sustained cooperation
CoalitionFormation-v0 6 Collective Action Coalition stability with exclusion
ApacheProject-v0 8-40 Validated (TR-3) Phase-dependent contributor dynamics
PublicGoods-v0 5 Collective Action Classic public goods contribution

TR-4: Sequential Interaction and Reciprocity (5 Environments)

Environment Agents Category Key Challenge
ReciprocalDilemma-v0 2 Reciprocity Conditional cooperation via memory
GiftExchange-v0 2 Reciprocity Asymmetric reciprocity sensitivity
IndirectReciprocity-v0 4 Reciprocity Reputation-mediated cooperation
GraduatedSanction-v0 6 Reciprocity Graduated proportional sanctions
AppleAppStore-v0 3 Validated (TR-4) Platform power and reciprocity dynamics

MARL Classification Summary

Quick reference for environment selection based on game-theoretic and MARL properties.

Game Type and Structure

Environment Game Type Cooperation Observability Symmetry
TrustDilemma-v0 Markov Game Mixed-Motive Full Symmetric
PartnerHoldUp-v0 Markov Game Mixed-Motive (power asymmetry) Full Asymmetric
PlatformEcosystem-v0 Markov Game / Mean-Field Hub-spoke topology Full Heterogeneous
DynamicPartnerSelection-v0 Markov Game / Mean-Field Reputation externalities Full Symmetric
RecoveryRace-v0 Markov Game Cooperative goal Full Symmetric
SynergySearch-v0 Bayes-Adaptive MG Unknown synergy Partial (hidden γ) Symmetric
SLCD-v0 Markov Game Mixed-Motive coopetition Full Near-symmetric
RenaultNissan-v0 Markov Game Phase-dependent Full Asymmetric
CooperativeNegotiation-v0 Markov Game + Contracts Enforceable agreements Full Symmetric
ReputationMarket-v0 Markov Game + Tiers Reputation competition Full Symmetric
TeamProduction-v0 N-player Markov Game Mixed-Motive (team vs. individual) Full Symmetric
LoyaltyTeam-v0 N-player Markov Game Mixed-Motive + loyalty amplification Full Symmetric
CoalitionFormation-v0 N-player Markov Game Coalition-based with exclusion Full Dynamic membership
ApacheProject-v0 N-player Markov Game Open source contribution Full Phase-symmetric
PublicGoods-v0 N-player Markov Game Classic public goods Full Symmetric
ReciprocalDilemma-v0 2-player Markov Game Mixed-Motive + reciprocity Full Symmetric
GiftExchange-v0 2-player Markov Game Asymmetric gift exchange Full Asymmetric
IndirectReciprocity-v0 4-player Markov Game Reputation externalities Full Symmetric
GraduatedSanction-v0 6-player Markov Game Common-pool resource Full Symmetric
AppleAppStore-v0 3-player Markov Game Platform coopetition Full Asymmetric

Action and State Spaces

Environment Action Space State Dim Horizon Early Termination
TrustDilemma-v0 Continuous [0,100]² 17 T=100 Trust collapse
PartnerHoldUp-v0 Continuous [0,120]×[0,80] 17 T=100 Weak partner exit
PlatformEcosystem-v0 Continuous [0,150]×[0,80]ᴺ (N+1)+3(N+1)²+1 T=100 Ecosystem death
DynamicPartnerSelection-v0 Continuous [0,100]ᴺ N+3N²+1+N T=50 None
RecoveryRace-v0 Continuous [0,100]² 17 T=150 Success/Collapse
SynergySearch-v0 Continuous [0,100]² 17 (or 18) T=100 Trust collapse
SLCD-v0 Continuous [0,100]² 17 T=100 Relationship breakdown
RenaultNissan-v0 Continuous [0,90]×[0,100] 17 T=100 Alliance dissolution
CooperativeNegotiation-v0 Continuous [0,100]² 18 T=100 Trust collapse
ReputationMarket-v0 Continuous [0,100]ᴺ N+3N²+1+N T=100 None
TeamProduction-v0 Continuous [0,50]ᴺ Phase-dependent T=100 None
LoyaltyTeam-v0 Continuous [0,50]ᴺ Phase-dependent T=100 None
CoalitionFormation-v0 Continuous [0,50]ᴺ Phase-dependent T=150 Coalition collapse
ApacheProject-v0 Continuous [0,50]ᴺ Phase-dependent T=60 None
PublicGoods-v0 Continuous [0,endowment]ᴺ Phase-dependent T=100 None
ReciprocalDilemma-v0 Continuous [0,100]² 17+ T=100 None
GiftExchange-v0 Continuous [0,100]×[0,80] 17+ T=100 None
IndirectReciprocity-v0 Continuous [0,100]⁴ Phase-dependent T=150 None
GraduatedSanction-v0 Continuous [0,100]⁶ Phase-dependent T=200 None
AppleAppStore-v0 Continuous [0,100]×[0,80]×[0,60] Phase-dependent T=66 None

Canonical Literature Comparisons

Environment Related Benchmarks & Literature
TrustDilemma-v0 Continuous IPD; Lerer & Peysakhovich (2017)
PartnerHoldUp-v0 Principal-Agent; Stackelberg games
PlatformEcosystem-v0 Mogul (ICML 2020); Multi-Principal Multi-Agent
DynamicPartnerSelection-v0 Resnick & Zeckhauser (2002); Rating systems
RecoveryRace-v0 Kim et al. (2004) trust repair
SynergySearch-v0 Bayes-Adaptive MDP; Duff (2002)
SLCD-v0 Ritala & Hurmelinna-Laukkanen (2009)
RenaultNissan-v0 Segrestin (2005) “Partnering to Explore”
CooperativeNegotiation-v0 Crawford & Sobel (1982); Raiffa (1982)
ReputationMarket-v0 Shapiro (1983); Tadelis (1999)
TeamProduction-v0 Holmström (1982); Alchian & Demsetz (1972)
LoyaltyTeam-v0 Akerlof & Kranton (2010); Kandel & Lazear (1992)
CoalitionFormation-v0 Ray (2007); Greenberg (1994)
ApacheProject-v0 Mockus et al. (2002); Lerner & Tirole (2002)
PublicGoods-v0 Fehr & Gächter (2000); Ledyard (1995)
ReciprocalDilemma-v0 Axelrod (1984); Killingback & Doebeli (2002)
GiftExchange-v0 Fehr, Kirchsteiger & Riedl (1993); Akerlof (1982)
IndirectReciprocity-v0 Nowak & Sigmund (1998, 2005); Panchanathan & Boyd (2004)
GraduatedSanction-v0 Ostrom (1990); Fehr & Gächter (2000)
AppleAppStore-v0 Parker, Van Alstyne & Choudary (2016); Rochet & Tirole (2003)

Special Features

Environment Distinguishing Mechanism
PartnerHoldUp-v0 Asymmetric interdependence (D=0.35 vs D=0.85)
PlatformEcosystem-v0 Hub-spoke topology, ecosystem collapse
DynamicPartnerSelection-v0 Public reputation signals
RecoveryRace-v0 Trust ceiling constraint (Θ = 1 - R)
SynergySearch-v0 Hidden complementarity parameter γ
SLCD-v0 Empirically validated (58/60 accuracy)
RenaultNissan-v0 Four configurable historical phases
CooperativeNegotiation-v0 Endogenous agreement formation, breach penalties
ReputationMarket-v0 Four-tier reward multipliers (0.40× to 1.30×)
TeamProduction-v0 Nash equilibrium baseline, free-rider dynamics
LoyaltyTeam-v0 TR-3 loyalty modifiers (φ_B=0.8, φ_C=0.3)
CoalitionFormation-v0 Dynamic exclusion/reentry, minimum coalition
ApacheProject-v0 Empirically validated (52/60), four project phases
PublicGoods-v0 Configurable multiplier, punishment mechanism
ReciprocalDilemma-v0 TR-4 reciprocity modifier, k=5 memory window
GiftExchange-v0 Asymmetric reciprocity (ρ worker 2.3× employer)
IndirectReciprocity-v0 4-agent reputation cascades, k=7 memory
GraduatedSanction-v0 Graduated sanctions (κ=0.8), k=10 escalation
AppleAppStore-v0 Empirically validated (48/55), 66-quarter history

Equilibrium Summary

Environment Stage-Game NE Pareto Optimal Price of Anarchy Notes
TrustDilemma-v0 $a^* \approx 35$ $a^* = 100$ ~1.55 Trust-mediated cooperation
PartnerHoldUp-v0 (45, 30) (120, 80) ~1.58 Asymmetric Stackelberg
PlatformEcosystem-v0 (55, 35) (120, 65) ~1.50 Collective action threshold
SynergySearch-v0 Conditional on $\gamma$ Conditional ~1.10 Bayesian exploration needed
SLCD-v0 Validated Validated N/A 58/60 historical accuracy
RecoveryRace-v0 Trust-constrained Recovery-dependent N/A Ceiling $\Theta = 1 - R$
CooperativeNegotiation-v0 Pre-agreement Post-agreement ~1.40 Breach penalty enforces
ReputationMarket-v0 Tier-dependent Premium tier ~1.35 Reputation competition
TeamProduction-v0 $a^* \approx 6.8$ $a^{opt} \approx 18.4$ ~2.5 Free-rider equilibrium
LoyaltyTeam-v0 Above Nash Social optimum ~1.2 Loyalty sustains cooperation
CoalitionFormation-v0 Coalition-stable Full coalition ~1.4 Exclusion threat maintains
ApacheProject-v0 Phase-specific Validated N/A 52/60 historical accuracy
PublicGoods-v0 Zero contribution Full contribution ~2.0 Classic public goods
ReciprocalDilemma-v0 ~35 (similar to TD) 100 ~1.55 Reciprocity enables TFT
GiftExchange-v0 Low effort from worker Fair exchange ~1.6 Asymmetric reciprocity
IndirectReciprocity-v0 Free-riding Full cooperation ~2.0 Reputation sustains cooperation
GraduatedSanction-v0 Under-contribution Full contribution ~2.0 Graduated sanctions deter
AppleAppStore-v0 Phase-specific Validated N/A 48/55 historical accuracy

Key Insights:

Categories

Dyadic Environments (2-Agent)

Micro-level scenarios modeling direct partnerships between two agents. Ideal for:

Environments:

Ecosystem Environments (N-Agent)

Macro-level scenarios with multiple interacting agents. Ideal for:

Environments:

Benchmark Environments

Research-focused environments designed for algorithm evaluation. Ideal for:

Environments:

Validated Case Studies

Environments with parameters validated against real business data. Ideal for:

Environments:

Extended Environments

Advanced scenarios with additional game mechanics. Ideal for:

Environments:

Collective Action Environments (TR-3)

Team production and collective action scenarios with loyalty dynamics. Ideal for:

Environments:

Reciprocity Environments (TR-4)

Sequential interaction and reciprocity scenarios with bounded memory. Ideal for:

Environments:

Common Interface

All environments share a common interface:

import coopetition_gym

# Create environment
env = coopetition_gym.make("EnvironmentName-v0")

# Reset
obs, info = env.reset(seed=42)

# Step
obs, rewards, terminated, truncated, info = env.step(actions)

# Access spaces
obs_space = env.observation_space
act_space = env.action_space

# Get info
n_agents = env.n_agents
endowments = env.endowments

Choosing an Environment

By Learning Challenge

Challenge Recommended Environment
Basic MARL TrustDilemma-v0
Credit assignment PlatformEcosystem-v0
Partner selection DynamicPartnerSelection-v0
Hidden states SynergySearch-v0
Long-term planning RecoveryRace-v0
Power asymmetry PartnerHoldUp-v0
Communication CooperativeNegotiation-v0
Market dynamics ReputationMarket-v0
Free-rider problems TeamProduction-v0
Loyalty dynamics LoyaltyTeam-v0
Coalition stability CoalitionFormation-v0
Empirical validation ApacheProject-v0, SLCD-v0, AppleAppStore-v0
Public goods PublicGoods-v0
Conditional cooperation ReciprocalDilemma-v0
Asymmetric reciprocity GiftExchange-v0
Reputation dynamics IndirectReciprocity-v0
Commons governance GraduatedSanction-v0
Platform reciprocity AppleAppStore-v0

By Research Area

Research Area Recommended Environments
Game Theory TrustDilemma-v0, SynergySearch-v0, TeamProduction-v0
Platform Economics PlatformEcosystem-v0, ReputationMarket-v0
Alliance Management SLCD-v0, RenaultNissan-v0
Trust & Reputation RecoveryRace-v0, DynamicPartnerSelection-v0
Negotiation CooperativeNegotiation-v0, PartnerHoldUp-v0
Collective Action TeamProduction-v0, LoyaltyTeam-v0, PublicGoods-v0
Coalition Theory CoalitionFormation-v0
Open Source Dynamics ApacheProject-v0
Mechanism Design LoyaltyTeam-v0, PublicGoods-v0
Reciprocity & Memory ReciprocalDilemma-v0, GiftExchange-v0
Reputation Systems IndirectReciprocity-v0, DynamicPartnerSelection-v0
Commons & Sanctions GraduatedSanction-v0, PublicGoods-v0
Platform Ecosystems AppleAppStore-v0, PlatformEcosystem-v0

Environment Comparison

Trust Dynamics Intensity

Environment Trust Sensitivity Reputation Effects
TrustDilemma-v0 High Moderate
PartnerHoldUp-v0 Very High High
RecoveryRace-v0 Extreme Very High
PlatformEcosystem-v0 Moderate Moderate
CooperativeNegotiation-v0 High High

Loyalty Dynamics Intensity (TR-3)

Environment Loyalty Sensitivity Free-Rider Pressure
TeamProduction-v0 None (baseline) Very High
LoyaltyTeam-v0 High Moderate (mitigated)
CoalitionFormation-v0 High High (exclusion threat)
ApacheProject-v0 Phase-dependent Phase-dependent
PublicGoods-v0 Moderate High

Reciprocity Dynamics Intensity (TR-4)

Environment Reciprocity Sensitivity Memory Length Agents
ReciprocalDilemma-v0 Moderate (ρ=0.5) k=5 2
GiftExchange-v0 High (asymmetric, ρ up to 0.7) k=3 2
IndirectReciprocity-v0 Moderate (ρ=0.32, but 3 partners) k=7 4
GraduatedSanction-v0 Graduated (ρ=0.12, but 5 partners) k=10 6
AppleAppStore-v0 Very High asymmetry (0.16-0.82) k=4 3

Scalability

Environment Fixed Agents Configurable Max Tested
TrustDilemma-v0 2 No 2
PlatformEcosystem-v0 1+N Yes 20
DynamicPartnerSelection-v0 N Yes 20
ReputationMarket-v0 N Yes 20
TeamProduction-v0 N Yes 20
LoyaltyTeam-v0 N Yes 20
CoalitionFormation-v0 N Yes 20
ApacheProject-v0 Phase-specific Yes (phase) 40
PublicGoods-v0 N Yes 20
ReciprocalDilemma-v0 2 No 2
GiftExchange-v0 2 No 2
IndirectReciprocity-v0 4 No 4
GraduatedSanction-v0 6 No 6
AppleAppStore-v0 3 No 3

Next Steps

Technical Reports