# Stored synthetic networks and preference rules

Run `python reproduce.py` with numpy and matplotlib. Standalone export/replot of a bundled neutral numeric snapshot: all24 network cells and five regional preference means, original publication-style figures PNG/SVG/PDF, runtime and exact source/generator hashes. No raw simulation or bootstrap is rerun. Originals are preserved. CSV values are copied exactly; only names/metadata are corrected.

## One-chance notice cascade

The original code allows newly informed nodes one opportunity per still-uninformed neighbor, then replaces the active frontier. Older nodes stay counted but do not retry failures. Correct public label: probability per eligible one-time attempt. This is not a repeated-transmission process. Under independent attempts from k newly active neighbors, the assigned activation probability is1-(1-p)^k. No belief, trust or real-world effect is modeled.

Stored setup:120 nodes, initial degree6,360 undirected edges; ring versus3,600 accepted degree-preserving double-edge swaps. This is not the standard small-world rewiring algorithm or a proven uniform graph sampler. Degree equality holds before edge deletion. Remove round(360*f) edges for f=0,0.1,0.3. Transmission p=0.1,0.3,0.6,0.9; at most20 rounds;20 blocks with50 cascades per cell/block,1,000 cascades per cell,24,000 total. Seed20260924. Each block's damaged graph is reused across the four probabilities. No-deletion ring blocks share the same topology; they are not20 distinct random ring graphs.

Coverage includes the seed node and is stored as a fraction. Intervals are stored percentile bootstrap bounds from5,000 resamples of20 block means, not intervals for a real population or individual cascades. Conditional time column is the mean of successful-run mean times within qualifying blocks, then averaged equally across those blocks. It is not pooled across all successful runs or unconditional. Empty CSV cell means no qualifying success, not zero time. The figure uses only coverage with its stored block intervals; reach90 and conditional time remain downloadable with these caveats. Raw individual trajectories and block arrays are not retained in this snapshot, so intervals cannot be independently recalculated from it.

## One-shot preference rule

Five regions each contain2,000 invented agents, with initial preference1 probabilities0.2,0.35,0.45,0.65,0.8. Observation coverage0.8, report-flip probability0.05, adoption probability0.5 when disagreeing. Seed20260925,100 trials. Compare one pooled observed majority against separate regional majorities; ties choose1. Adoption draws differ between the two policies. No repeated feedback, delay or dynamic stability is implemented.

Mean match with INITIAL preference: pooled0.754932; regional0.844829. Doing nothing scores1 by definition. This metric penalizes changes and is not welfare, learning, fairness, legitimacy or policy quality. Figure shows stored regional means with no invented confidence intervals; raw trials were not retained. The equality of the global mean and mean of five region means is checked because all regions have equal size. Neither rule's results establish superiority for real communities. Zero observation coverage is outside this stored setup; the original tie rule should not be interpreted as evidence-based estimation without observations.

Checks verify all24 unique cells, bounds, interval ordering, missing-time semantics and global/region consistency. Exact raw-result comparison is documented separately; this export is not empirical or independent experimental replication. Primary definition of the one-chance mechanism: Kempe, Kleinberg and Tardos(2003), https://www.cs.cornell.edu/home/kleinber/kdd03-inf.pdf . This reference does not validate the assigned network or social claims.