Week 1: One System, Two Stories
The opening lecture — the same synthetic system described by an annual figure and by its hourly trace, and why only one of them is a design input.
27 July 2026
You will be handed a system described by one number and asked whether it is reliable. In this course the number is 87.4% renewable across the year, a property of the generated teaching trace rather than a measurement of anything. It is not a lie. It accurately reports what it was built to report. The question this lecture puts in front of you is narrower and harder: does it report anything you can sign your name to?
An annual share is a sum divided by a sum#
The operation that produces an annual figure is addition. Every hour of generation is added together, every hour of demand is added together, and one total is divided by the other. Addition is commutative — it does not care what order the hours arrived in. That is exactly the property that makes the result useless for reliability, because reliability is entirely a question about order. A system that generates its shortfall as one hour a week survives it. The same system generating the same annual shortfall as one continuous stretch does not.
So an annual share is not a weak reliability statistic. It is not a reliability statistic at all. It answers a question about energy accounting, and it answers it correctly, and reliability is a different question about the same data.
What the summary discards#
Read the course’s synthetic fortnight two ways. Across all 336 hours the combined wind and solar output averages 27.2% of installed capacity. Inside the 61-hour reference event it averages 3.6%, and at its lowest hour it reaches 0.9%. Eighty-eight of the 336 hours sit below the course’s 12% threshold. Every one of those figures is a property of the generated trace or a boundary this course chose, and the fortnight is labelled synthetic wherever it appears.
Now notice what a reader given only the mean can reconstruct from it. Not the event’s length. Not its depth. Not whether there was an event at all. The mean is not a compressed version of the trace with some detail lost; it is a different object, from which the trace cannot be recovered even in outline. The information was not degraded. It was destroyed.
The tail is not an outlier here#
The reflex when this is pointed out is to reach for a worst-case number instead: report the minimum. That reflex is the misconception this lecture exists to break.
The minimum is one hour. A system rides through one bad hour on inertia, reserves and the fact that nothing has had time to drain. The 0.9% hour in the trace is not what threatens the system; the 61 hours around it are. Neither the average nor the extreme is the right statistic, because the thing that defeats a renewable system is not a value at all. It is an interval — a stretch of time of some length, below some level, which is a shape and not a point.
This is why the tail matters here in a way it does not in most engineering statistics. An outlier is a rare value you can bound and move on from. A drought is not a rare value; it is an ordinary value held for an unusual length of time. Every hour inside the reference event would be unremarkable on its own.
What this changes about the rest of the semester#
If the object of study is an interval rather than a value, then before anything can be designed, predicted or defended, the interval has to be specified. How low, for how long, across which sources, with how much recovery permitted inside it before it counts as two events instead of one. None of those are observations. They are decisions, and different decisions produce different answers from identical weather.
That is the subject of week 2 and the reason the course fixes one operative definition at the event instead of treating “drought” as self-explanatory. It is also why the first assessment, Define the Drought, asks you to move that definition and defend the move rather than accept the one you were given.
For this week, the discipline is narrower. In the studio, week 1 asks you to find the worst window in the trace yourself, and to say what makes it the worst — which forces you to choose a length before you can rank anything. Bring the answer you actually derived, not the one that sounded robust. Every number you meet in this course carries a label saying whether it was chosen, generated or measured. Start reading those labels now; by week 12 you will be expected to defend which kind each of your own numbers is.