Chapter 5 · Tool twoBuild it to bend

You cannot predict the shock. You can build the thing that takes it and keeps going. A resilient system does not need to see the future — it is shaped so the future cannot break it. Four moves, each one drawn.

Move one — pull back to centre

The oldest trick in engineering: measure how far off you are and push back a share of it, every tick. A woodstove with a thermostat. A hand on a wheel on a washboard road. James Clerk Maxwell wrote the mathematics of it in 1868 and started a whole field.²⁰

Fig 10. Two tanks take the very same run of random shocks. One just piles them up and wanders off (it drifted to 6.92 from centre). The other subtracts a fraction of its own error every step and stays put (it held to 0.61). Same shocks. One rule of feedback between them. · computed here · Nan · hongdam.net · CC BY 4.0

error_next = (1 − k) · error_now

Reading it: each step, keep only the fraction (1 − k) of the error you had. As long as the pull-back k is between 0 and 2, the error dies away on its own. Pull too hard, past 2, and you overshoot worse each time — feedback wired backwards makes its own chaos.

Move two — keep spares

If one part fails one time in ten, two that fail on their own both fail only one time in a hundred; three, one in a thousand. Independence is a fierce multiplier — if you can keep it.

Fig 11. The straight drop is the dream: each spare cuts the failure rate ten-fold. The two dashed lines are the catch — let a common cause reach across your spares (one power feed, one supplier, one bad assumption) and the floor stops falling. Redundancy is only worth what its independence is worth. · computed here · Nan · hongdam.net · CC BY 4.0

P(all N fail) = p^N (only if they fail independently)

Reading it: the chance every one of N parts fails at once is the single-part chance multiplied by itself N times — a tiny number, fast. The parenthesis is the whole ballgame: shared causes break the multiplication.

Move three — spread the bet

This is the law of large numbers from chapter four, put to work on purpose. Split a stake across many independent bets and the swing shrinks as one over the square root of their number — same expected return, less lurch. Harry Markowitz won a Nobel for writing this down plainly in 1952.¹⁸

Fig 12. Spread across independent bets and the swing keeps falling. But the moment the bets share a common risk — one market, one weather, one rumour — the fall stops at a floor no further spreading can get under. That floor is the shared risk, and only a different kind of bet removes it. · computed here · Nan · hongdam.net · CC BY 4.0

Move four — stay off the edge

Chapter two showed a wall where a steady system tips into chaos, with the forks crowding up to it. A resilient system is run with room to that wall, not parked against it for the last drop of yield. C. S. Holling, studying forests and fisheries, drew the line between two things people muddle: how fast a system springs back, and how big a shock it can take before it flips to a different world entirely.¹⁹ Slack is what buys the second one. A system with no slack is fast and efficient right up until the day it is neither.

The tool, in one line each

Pull back to centre — wire in something that corrects error on its own.
Keep spares — and guard their independence harder than the spares themselves.
Spread the bet — many small uncorrelated exposures beat one big right guess.
Stay off the edge — keep margin to the tipping point; measure the shock you can take, not only the speed you bounce back.

None of these needs a forecast. That is the point. You are not trying to out-guess the chaos. You are building something that does not need to.

← Tool one: large numbers Do it yourself →