Margins, Reserves, and the Illusion of Enough

Every grid operator will tell you they hold a reserve margin — capacity above expected peak demand, kept in hand for the bad day. The number sounds reassuring. It is also one of the most misunderstood figures in infrastructure, and misunderstanding it is how competent organizations walk into blackouts with a clean conscience.

The trouble starts with how the margin is calculated. Reserve margin is nameplate capacity minus expected peak, expressed as a percentage. It quietly assumes that capacity is available when called. But capacity on paper is not power on the wire. A generator down for maintenance contributes nothing. A wind fleet contributes what the wind allows. A gas plant contributes only if the fuel arrives. The margin, in other words, is an accounting figure, and the grid does not run on accounting. It runs on what physically shows up in the hour of need — and the hour of need is precisely the hour when the fewest things show up, because the conditions that drive demand to its peak are often the same conditions degrading supply.

This is the heart of the matter, and we will return to it repeatedly across this course: correlation. A reserve margin is only meaningful if the failures it guards against are independent — if the probability of any one unit being unavailable is unrelated to any other. Under that assumption, holding a fifteen percent margin against a fleet of many units is genuinely robust, because the odds of many failing at once are the product of small numbers, which is a very small number. The mathematics is comforting and the mathematics is a trap. In the real world the failures are not independent. A cold snap that spikes heating demand also freezes wellheads, ices wind turbines, and strands the gas that three different plants were all counting on. One weather system, many simultaneous failures, all of them correlated through a common cause. The margin that looked like fifteen percent on a spreadsheet was, on that morning, something closer to zero, because the spreadsheet had modeled the units as independent and the weather had not read the spreadsheet.

There is a name for the discipline that tries to fix this — resource adequacy — and it has grown steadily more honest over the years, moving from simple margin percentages toward probabilistic measures that ask how often, and how deeply, the system is expected to fall short. These loss-of-load metrics are a real improvement because they force you to reason about the tail of the distribution rather than the average. But they are only as good as the correlations baked into them, and correlations are exactly what organizations are worst at estimating. It is easy to model each generator's failure rate. It is hard, and unglamorous, and politically inconvenient to model the day they all fail together — because that model tells you your comfortable margin is a fiction, and someone will have to pay to fix it.

Consider what 'holding reserve' even means in practice. Spinning reserve is generation already synchronized to the grid, turning, ready to ramp within seconds — expensive, because you are burning something to produce power nobody is yet buying. Non-spinning reserve is capacity that can be brought online within minutes, cheaper but slower. Operating reserves are the fast twitch of the system, sized to cover the sudden loss of the single largest unit. But none of these help against a slow, broad, correlated squeeze where demand climbs for hours while supply quietly erodes across the whole fleet at once. That is not a twitch. That is a tide coming in, and reserves sized for twitches drown in tides.

The practitioner's lesson is to treat every stated margin as a question rather than an answer. Ask what it assumes about availability. Ask whether the failures it guards against are independent or merely modeled as independent. Ask what single common cause could remove a large fraction of the fleet at once, and whether the margin survives that scenario. The margin that only holds under independence is not a margin; it is a hope with a percentage sign attached.

Faraday's read: 'enough capacity' is the most dangerous phrase in grid planning, because it is almost always true on average and almost always false in the moment that matters. Adequacy is not a stockpile. It is a bet about correlation, and the house — the weather — knows the odds better than you do.