What a Global Technology Leader Taught Us About Precision.

01Where the unit of argument is very small

In advanced technology environments, tolerance is not a preference expressed by a demanding customer. It is physics. A dimension is either inside the window or the part does not do its job, and no amount of good intent moves the window. The first thing you learn is that the seriousness in the room is not corporate temperament — it is a rational response to what the process actually requires.

The second thing you learn is how little of that seriousness is about heroics. It is almost entirely about method: measure, record, check against the specification, and treat the check as real work rather than an obligation to be compressed when the schedule tightens.

02The document is part of the deliverable

Software culture tends to treat documentation as something written afterwards by whoever has time. In precision engineering it is the opposite: the record is the thing that makes the work usable by anyone other than its author. A part with no traceable history is not a finished part, whatever its dimensions.

That inverted the way our teams work. Design decisions are recorded when they are made rather than reconstructed later. Revisions are controlled, and a change nobody can point to is treated as a defect in its own right. It is slower on the day and faster over a year, which is a trade almost every engineering organisation says it believes in and rather fewer actually make.

A result nobody can reproduce is an anecdote. The record is what turns it into engineering.

03Small details, compounded

The habit that transferred furthest is the smallest one: taking minor imprecision seriously. A label that does not match the drawing. A unit left off a value. A naming convention broken in one file. Individually each is trivial and each has an obvious explanation, which is exactly why they survive.

In a demanding environment they do not survive, because everyone has seen what happens when a hundred of them accumulate in the same system. The engineer who chases the mismatched label is not being pedantic — they are refusing to let a system become one where nothing quite lines up and nobody can say which mismatch is the dangerous one.

• 01Units, always. On every value, in every document.
• 02The name in the drawing, the code and the report is the same name.
• 03A tolerance without a measurement method is not a tolerance.
• 04If two documents disagree, work stops until one of them is wrong.

04What we carried into everything else

It would be easy to file this as something that belongs to hardware. It does not. The same discipline is what makes an AI system defensible: knowing which version of the data trained which version of the model, recording why a threshold was set where it was set, and being able to answer 'what changed?' without a forensic exercise.

Precision is not perfectionism, and it is not slowness. It is the decision to be exact about the small number of things that determine whether everything else can be trusted — and it is the reason a system built years ago can still be reasoned about today.