Reproducing the phase-diagram paper
Target paper: A Phase Diagram of Decentralized Coordination Under Communication Delay.
Versions pinned for the canonical results
| Component | Version | SHA256 (short) |
|---|---|---|
gossamer_threaded_intelligence wheel | 0.3.1 | TBD |
leviathan (engine) | py-0.2.1 | TBD |
maneuver-map backend | TBD (git) | TBD |
| Python | 3.10.x | — |
Fill the TBD hashes from the provenance.packages block of any cell’s
committed experiment.json (the run recorded gossamer 0.3.1).
Environment setup
python3.10 -m venv .venv && source .venv/bin/activate
pip install --extra-index-url https://us-central1-python.pkg.dev/arboria-research/python-packages/simple/ \
gossamer-threaded-intelligence==0.3.1
export ENGINE_MODE=inprocess
export PYTHONHASHSEED=0The grid
The canonical run is the dcc_p1 batch: primitive × task × delay × N × seed.
| Axis | Values |
|---|---|
| primitive | no_comm, flocking, dmb_tf_aco, gossip, iccd |
| task | rendezvous, consensus, coverage_hold* |
| delay (ms) | 0, 5000, 10000, 20000, 40000, 60000 (= 0–60 steps at dt=1) |
| N | 500, 2000 |
| seed | 1, 2, 3, 4, 5 |
Total: 900 cells. *coverage_hold is collected but excluded from the
headline figures (no current primitive achieves targeted coverage — see the
paper’s scope section); dmb_tf_aco is a dispersal primitive and is omitted
from the collapse panels.
Fixed parameters: dt=1.0, tau_sec=300, bound=1000, init_spread=500,
comm_range=1200 (connected interaction graph), energy_rate=0 and
fault_prob=0 (so agent attrition cannot confound the delay axis),
integrator=velocity_verlet, steps=1500.
Reproduce
The grid was scheduled through Maneuver.Map’s /jobs/experiment sweep endpoint
(one Cloud Run task per cell); the same specs run locally via
scripts/run_paper_experiments.py --plan p1. Each cell writes an
experiment.json with the full provenance block. Figures regenerate from the
committed batch folder (dcc_p1-…) by aggregating task_metrics.Q_final over
seeds by (task, algorithm, delay, N).
Anchor
The Vicsek substrate anchor is the vicsek batch: flocking→vicsek primitive,
velocity_noise ∈ {0, 0.5, …, 5} × 5 seeds (45 cells), energy_rate=0. It
reproduces the order–disorder decay ψ: 1.00 → 0.63.
Expected outcome
no_comm holds the floor at every delay; the three communicating primitives
collapse together from Q ≈ 1 (delay 0) through a boundary near 10–20 steps to the
floor by delay 40, identically at N = 500 and N = 2000. The run produced 0
non-finite values and 0 agent faults.