The Closer You Look The Less Precise It Gets

Happiness is finding a new podcast to follow. This happened most recently when Roman Mars’ “99% Invisible” cross-posted a Hyperfixed episode about supermarket cold cases. I could tell pretty quickly that this would likely end up in the regular rotation and this morning found myself enjoying the latest episode on measurement.

This particular episode centers on a trip down a rabbit hole that starts with a listener wondering about folklore regarding the tab found on the end of a tape measure and ends up in some deeply philosophical territory with Mythbusters’ Adam Savage about precision (my own post’s title being a quote stolen from him). As the episode unfolded and then lingered in my mind I found myself reflecting on my own various experiences spanning robotics, locks, and software.

Most viscerally, the discussion of how a tape measure’s tab affects precision had me recalling a day about fifteen years ago when I realized belatedly that I had not thought adequately about whether I should be running a bandsaw through or next to a line I had drawn on bars of aluminum. I suppose you can do either but what is non-negotiable is that you pick an approach and stick with it if you’re ever going to get where you are trying to go. That particular sub-misadventure imbued in me a deep appreciation for how expensive mistakes in meatspace are relative to cyberspace on even the most trivial seeming matters as I pondered a pile of parts that represented an afternoon’s work, wasted because I could not bolt them together.

Within the episode you will find discussion of tools to perform measurements with far more hardcore precision than a tape measure, calipers being a notable example, a tool I did not own personally until I went down the robotics rabbit hole. At the most extreme there was discussion of means to measure down to the micron level with the unsettling nuance that at such precision it is now incumbent on you to mind the temperature if you want your tool to be of any real use to you. From here I found my brain moving laterally to the sporadic hobbyist lock picking I have done over the years, an exploration spawned long ago at DEFCON’s lock picking village, a realm where the exploits you execute hinge on imperceptibly small flaws in the milling of components that derive from drill bit wear which allow you to manipulate alignment and friction issues to collapse a theoretically exponentially complex problem space into a practically linear one (if only you can coax your fingers to such a degree of finesse).

Leaving the realm of my hobbies and focusing on a longtime professional reality, I found myself pondering high volume distributed systems, a realm where the non-practioner’s intuition on acceptable error rates generally lands several orders of magnitude wide of the mark. I remember sitting in a graduate level course run by Yair Amir some twenty years ago wherein he opined on how many “nines” of reliability you need when you put code into contexts where zillions of events happen per day and you care about all of them. This would only become more pressing during the intervening decades as we lived through the take-off of virtualization, containerization, and cloud computing. How many minute details beyond the ken of mere mortals would AWS need to nail to get to the “eleven nines” of S3’s durability guarantees?

There are worlds within worlds within worlds and you can spend a lifetime exploring a problem space without ever getting to the bottom it.