Log inGet started
World

Cascade — a self-organising system you have to keep safe

Branchmain
Branchesmain
Commit1e11026548

README.md

CascadeAgents — Phase 25, non-human baselines

Five strategies, from doing nothing to actively hunting the largest connected region. Each pays the same energy costs a player does, so the comparison is between strategies and not between budgets.

idnamewhat it does
do_nothingDo Nothingnever intervenes — the control condition
randomRandom Interventionacts at random whenever it can afford to
density_thresholdDensity Thresholdfires a controlled burn once occupancy passes a threshold
largest_clusterLargest Clusterburns the centroid of the largest connected region
firebreakFirebreakcuts a line across any region that grows too large
local Agents = require("CascadeAgents")
local rows = Agents.runAll({ seed = 20250922, width = 128, height = 128, ticks = 700 })
print(Agents.formatComparison(rows))

Headless

These run through Core.runHeadless — no rendering, no UI, no frames — so a strategy can be measured far faster than it can be watched.

Agents.makeRunner(id, opts) returns an incremental runner (stepChunk(n), progress(), finish()), and Agents.runAllChunked drives one for every strategy while yielding, which is what the instrument's B key uses. A long baseline run therefore happens behind a board that keeps moving.

Agents.largestRegionInfo(sim) gives a region's centroid and bounding box; it costs one pass over the board, so it is asked for on the agent's decision cadence and never per tick.