Principles
How we hold ourselves to the work.
Eight commitments we return to when the work gets hard — or when the results get interesting. Each carries the operational test that shows whether we actually kept it.
Method is not only technique; it is discipline. These are the standards we would want anyone to hold us to, and the ones we try to keep when nobody is checking. None of them is novel. Taken together they are what separates understanding from a confident guess.
Working principles
- 01
Wonder deserves discipline
Interest in a problem is the starting point, not the method. We give curiosity a structure — observe, measure, model, test — so it produces something more durable than an opinion.
Test: is there a written question, or only an enthusiasm?
- 02
Assumptions are stated, then tested
Every model rests on assumptions. We write them where they can be seen, questioned, and — when they fail — replaced. An unstated assumption is not a simplification; it is a hiding place.
Test: can a reader list what the model takes for granted?
- 03
Evidence outranks intuition
Intuition is a good source of hypotheses and a poor source of conclusions. When the two disagree, we go back to the measurement rather than to the argument.
Test: which measurement would change our mind?
- 04
Models are tools, not truth
A model is a useful compression of reality, not reality itself. We use models to reason and to predict, and we stay alert to the conditions under which they stop describing the world.
Test: where is this model known to break?
- 05
Uncertainty is measured, not concealed
Uncertainty is not a weakness in a result. Unmeasured uncertainty is. We try to say how confident we are, and on what basis.
Test: does the number come with an interval?
- 06
Methods should be inspectable
A result you cannot inspect is a result you cannot trust. We aim to make reasoning, data, and code legible to someone who wants to check our work rather than take it.
Test: could a stranger reproduce this from what we published?
- 07
Conclusions move with the evidence
Holding a position after the evidence has moved is not conviction. Revising a conclusion is the process working, not the process failing.
Test: is the record of what we changed still visible?
- 08
Understanding over speculation
It is easy to be interesting and wrong. We would rather understand a small thing well than narrate a large thing loosely — and say plainly which one we have done.
Test: can we predict something we have not already seen?
Uncertainty is not a weakness in a model.
Unmeasured uncertainty is.