Eco 2.1: Earned Boundary, Honest Detector

Provenance-gated ignition, per-patch, with a fatal ground-truth check

The successor to Eco 2.0, whose detector could be fooled by a handed-in boundary. Here every wall cell carries an immutable origin, so a handed-in boundary is inexpressible. But the first run exposed a second failure: with predators starved out early, walls piled up to 70% density and the “patches” were percolation pockets — an edge, not a defended boundary. This version is staged, and here is where it stands. Step 1 (walls off) passed: the bare tri-trophic loop coexists in a bounded oscillation for 11,000+ tiks. Step 2 (first attempt, walls cheap at 0.8) failed, controlled against step 1: herbivores walled the whole grid in ~20 tiks, predators starved by tik ~164, and the “patches” were percolation pockets — an edge, not a boundary. The diagnosis was the wall economy: a wall cost less than a fifth of a meal, so nothing had to be earned. This build raises the wall cost to a real sacrifice (3.5, more than a full meal), because a sovereign boundary must run at cost — that is Condition 4, not a tuning knob. The question now: with a costly boundary, does anything close a ring on purpose rather than by saturation, and do predators survive? The CSV now logs the build cost and decay on every row, so the run is reconstructable. Watch the predator count and the wall-cell count together.