When the work says stop
The SpaceX compensation package ties Musk's pay to a Mars colony of a million people and orbital data centers drawing a hundred terawatts - targets so far beyond current capability that calling them ambitious undersells the gap.1
The board signed them because they had watched him make predictions that sounded equally impossible for two decades and then deliver every one: private human spaceflight, reusable orbital rockets, a satellite internet business worth billions when the prior model was a graveyard of bankruptcies.
The scale of what he attempts and then achieves is the evidence I lean on when my own work feels impossible - not because I am building a Mars colony, but because it proves that "impossible" is a temporary label one person's sustained effort can remove. If a thirty-year-old with a spreadsheet of rocket material costs can found a company that now launches 83% of the mass sent to orbit, then the constraints I take as fixed are probably not.
When you work on something hard, the daily evidence is all friction: the thing that broke, the timeline that slipped, the person who said no. The signal you get from the work itself tells you to stop, and you need a counterweight that says keep going - a piece of evidence that a human being can reach a target every reasonable observer would call unreachable.
The SpaceX story supplies that counterweight in quantity. Falcon 1 survived three failures to make orbit on its fourth and final attempt with weeks of cash left. The Raptor engine was redesigned from scratch three times in under a decade while every other engine program iterates once a generation. A 100,000-GPU supercomputer was built in 122 days when the industry standard is four years.
Each of those is a data point that says your ceiling is wrong. Believing the ceiling is wrong is not naive optimism - it is the only position consistent with the evidence, because there is now a long record of one person reaching past ceilings that the rest of the world treated as walls.