nx_substrate_compose_test.nx

nx_substrate_compose_test.nx -- end-to-end compose demo. Exercises ALL 11 optimization-layer primitives together to verify they compose. Per cardinal real-playtest-loop, the per-primitive smokes don't catch composition bugs (one primitive can pass its own test while violating an invariant another primitive depends on). This smoke is the substrate's whole-cloth verification. Scenario: an "iridium voxel world" cell hosts a 32x32x32 tissue at 3 bits per voxel (12 KiB = ESP32-feasible). The cell declares: - foreground game attention class (frame budget pressure) - LOCAL_GPU tropism preference - PTX ribosome target - 16 KiB RAM ceiling (tissue + some scratch) We then simulate the lifecycle: 1. Construct the tissue, fill it half-air half-stone 2. Wrap the cell in a pathway 3. Profile a "raycast" call site through metabolism 4. Apply budget pressure 5. Ask homeostasis for a signal 6. Resolve tropism for the signal 7. Pick ribosome target for the destination tier 8. Log eviction journal entry for the migration Every step asserts the expected outcome from a foreground-game cell under realistic resource pressure.