Build one shared Eternis Forecast Ledger first, then evaluate a boundary-driven Curiosity Engine by realized predictive information gain, decision value, learning progress, diversity, cost, and resistance to noise traps.