Proposal: maintenance commons substrate for shared infrastructure cooperation

I'd like to propose a new Melting Pot substrate focused on cooperation around **degrading shared infrastructure**.

The core social dilemma would be that all agents benefit from infrastructure remaining operational, while maintenance has an individual opportunity cost. Agents can therefore maintain, free-ride, condition their contribution on others, or exploit periods of high infrastructure health.

A first version could include:

- a shared infrastructure health state that decays over time;
- maintenance actions that restore health at an individual cost;
- collective productivity/reward that depends on infrastructure health;
- heterogeneous roles or maintenance costs to create division-of-labour and fairness pressures;
- evaluation variants with unfamiliar partners and different mixtures of cooperative, conditional, and free-riding background policies.

The main research question would be whether agents generalize maintenance/cooperation strategies to unfamiliar partners and changed social compositions, rather than simply learning a fixed contribution convention.

This seems distinct from existing commons-harvest and factory substrates because the shared resource is not primarily depleted by consumption; instead, agents face a recurring costly-upkeep problem where neglect degrades the environment for everyone.

If this direction is in scope for Melting Pot, I would be happy to prototype the substrate, human-play/debug path, tests, documentation, and a small set of scripted/Puppeteer policies and evaluation scenarios. Before implementing it, I'd appreciate maintainer feedback on whether this is a useful social interaction to add and whether there are design constraints I should follow.

Start by reviewing the existing commons-harvest and factory substrates to compare their design constraints with the proposed maintenance substrate. After maintainers confirm scope, a complete contribution would cover infrastructure decay and maintenance costs, productivity rewards, heterogeneous roles, partner-generalization evaluations, human-play/debug support, tests, documentation, scripted/Puppeteer policies, and evaluation scenarios.

Written by the indexing model from the issue text.