Llm Coordination Harness

Llm coordination harness

[](./pyproject.toml) [](./docs/CLAIM.md) [](./docs/RUNBOOK.md) [](./docs/TECHNICAL_REPORT_DRAFT.md) [](./configs/RELEASE_FREEZE.json)

`llm-coordination-harness` is a reproducible measurement rig for hidden coordination variables in multi-agent LLM systems under fixed billed-token budgets.

This repository is intentionally positioned as:

• an eval / harness project
• a measurement-first research artifact
• a negative-results / methods result

This repository is not positioned as:

• a generic swarm framework
• a production routing layer
• a claim that a universal coordination law has already been proved

What It Measures

The harness extracts and logs:

• F: critical-fact survival fidelity through the graph
• rho: shared-error correlation under no communication
• B: propagation balance over edge fact-survival ratios
• C: fan-in pressure from incoming peer-token load vs quota

The important property is that these variables are recomputed from logs offline, rather than living only in process memory.

Golden Artifacts

The `outputs/` directory is intentionally frozen to the two gold runs:

• p0a-calibrated-full-live
• p0b-attacks-live

Frozen configs:

• p0a_calibrated_full_live.yaml
• p0b_attacks_live.yaml
• RELEASE_FREEZE.json

Visuals

Feature importance from the offline predictor analysis:


Topology penalt