CI Library

How the library grows

Six independent tracks, one loop from theory to practice.

The Collective Intelligence Library is in its construction and simulation phase. The work runs as six tracks. Each track moves on its own, and each track can take outside contributions on its own.

Follow the tracks ↓

Phase
Construction & Simulation
Tracks
6
The two phases
Phase 01 / Now

Construction & Simulation

Build the simulation infrastructure and test defences inside it. The six tracks below all belong to this phase.

followed by
Phase 02 / After

Practice

Test the policies and defences that win in simulation with real organisations, in the real world.

The six tracks

Each track answers one question and moves through its own milestones. Together they turn the library from an engine into shared infrastructure.

Milestone states
Done
Active
Next
Later
Track 01

Metrics & Coupling

What should we measure, and how do systems connect?

  1. Two-layer metric frameworkDone
  2. Cross-disciplinary catalogActive
  3. Calibration against known sourcesNext
  4. Multi-domain coupling contractNext
  5. Coupled system measured end to endLater
Track 02

Scenarios

Which failures should we be able to rehearse?

  1. First scenario benchmarkedDone
  2. Detailed economic modelActive
  3. Political & cultural modelsActive
  4. Coupled disempowerment suiteNext
  5. AI epistemics scenarioNext
  6. Runnable canonical futuresLater
Track 03

Mechanisms

Which parts can the field share?

  1. Typed read/write declarationsDone
  2. Markets, voting, networksActive
  3. Imports from more disciplinesNext
  4. Contribution pathwayNext
  5. First external module mergedLater
Track 04

Interface & Interoperability

Can people outside the code see the model?

  1. Live showcaseDone
  2. Assumption-inspection viewsActive
  3. Machine-readable scenario pagesNext
  4. Interop with other testbedsLater
Track 05

Agents

Who can live inside a scenario?

  1. Rule-based agentsDone
  2. Learning agentsActive
  3. Active inference agentsActive
  4. Deliberative reasoning agentsNext
  5. Language-model agentsLater
  6. Large-scale LLM populationsLater
Track 06

Performance & Scale

How large can a population get?

  1. Seed-parallel runsDone
  2. Population-scale profilingNext
  3. Parallel engine paths in JAXNext
  4. 10,000+ agents on one GPULater
Where it ends

Closing the loop

All six tracks belong to the construction and simulation phase. The phase after it is practice: testing the policies and defences that win in simulation with real organisations. That closes the loop between theory and practice, and it is the point of the whole pipeline.