Practice

Take two pieces of cast iron, each machined flat to the best tolerance the machine can hold, and put one on top of the other. They will not touch across their faces. They will touch in three or four places, and the rest will be air.

The way you find out where is to paint one surface with a thin blue marking compound, press the two together, separate them, and look. The blue transfers only at the high points. What you are holding is a map, and it is a map of something no drawing contains, which is where these two particular surfaces actually meet.

Then you take a hand scraper and remove a few microns from each blue spot. Re-blue. Press. Look again. The contact spreads. Repeat, for hours, and the two surfaces converge on a fit that is more accurate than the machine that made them, and more accurate than anything either surface started out as.

Three things about that process are the whole of this chapter.

The map cannot be predicted. Nobody can tell you in advance where the blue will show, because it depends on this casting, cooled this way, machined on this machine, on that day. The general theory of flatness is well understood and it will not tell you which spots to scrape.

The accuracy is generated rather than transferred. A scraped surface can be flatter than the reference it was made against, which is why the technique underpins precision engineering rather than merely copying it. The skill produces something the inputs did not contain.

And it cannot be bought. The scraper, the blue and the granite reference cost very little compared with the machine tool. What you cannot buy is the hand that knows how hard to press and how much to take off, and that hand is made by scraping, badly at first, on real work, under somebody who has done it.

Practice is the second discipline in the Process layer. Permission decided what the system may reach. Practice is what happens when people who have to get work done sit down with it, and it is the activity of finding out where the tool actually touches the work rather than where the diagram says it should.

In 2026 the strongest available evidence for that claim stopped being anecdotal, because somebody ran the experiment.