Sizing against the duck

Every kilowatt-hour you generate on site displaces a grid kilowatt-hour, but not an average one. It displaces the specific hour's kilowatt-hour, and in California the hours are no longer alike. Sizing that ignores the shape is arithmetic on a grid that no longer exists.

Section 01What the duck actually is

Plot California's demand net of wind and solar across a spring day and the line sags deep in the midday hours, then climbs steeply into the evening as the sun sets while demand holds: a silhouette the industry has called the duck curve since the California Independent System Operator and national laboratories first documented it.¹ The U.S. Energy Information Administration's follow-on analysis is blunt about the trend: as solar capacity has grown, the midday belly has dropped lower year over year, and the evening ramp, the duck's neck, has steepened.²

Prices follow the shape. Wholesale electricity in California is now systematically cheapest in the solar hours and most expensive at the morning and evening shoulders, a pattern EIA documented as the inversion of the old midday-peak world.³ Retail time-of-use tariffs translate the same shape to customers: the expensive evening window, the cheap midday window, and rate designs that keep migrating the "peak" label later into the day.

Section 02The storage fleet is reshaping the duck in real time

The state's response has been the fastest battery build-out in the country: California surpassed 21,000 megawatts of battery resources in 2026 by the state's own accounting, charging through the cheap midday belly and discharging into the evening neck.⁴ CAISO's public dashboards now show storage as a routine top-tier participant in the evening supply stack.⁵

For a behind-the-meter planner this matters in two directions. It means the extreme evening scarcity that made simple arbitrage look lucrative is being competed away at grid scale, so a customer-sited battery's value case must rest on the tariff and demand charges actually in front of it, not on wholesale folklore. And it means the shape itself will keep evolving; a sizing decision tuned precisely to 2026's curve is tuned to a moving target. Robustness beats optimization.

Design against the tariff you are actually on, with the load you actually have. The duck is context; your meter is the contract.

Section 03What the shape does to each sizing strategy

Section 04An illustration, labeled as one

Consider, purely as an illustrative example, a facility whose time-of-use tariff prices the midday window at roughly half its evening window. A flat around-the-clock generator displaces a blend of both and earns the blend. A generator or battery that concentrates the same annual energy into the expensive windows earns materially more per unit, at the cost of more complex operation, more cycling wear, and dependence on tariff windows that the utility can and does revise. Neither answer is universally right; the point is that the spread between them is now large enough that a study which models one flat avoided rate has skipped the actual decision. Real analysis runs your interval data against your filed tariff, hour by hour, for a full year, and then stress-tests the answer against a plausible future tariff revision.

The 8,760-hour method, step by step

The analysis this paper keeps invoking is mechanical enough to describe completely. First, obtain a full year of your own interval data, ideally at fifteen-minute resolution, through your utility's data-access channel. Second, obtain your actual filed tariff, not a summary, and map its price windows onto every hour of that year: which hours bill at which energy rate, where the demand-charge measurement windows sit, and how seasons shift the map. Third, for each candidate design, simulate its hourly behavior against your load: a flat generator runs flat; a battery charges and discharges within its capacity, its duration, and its cycling limits; a shaped machine follows the dispatch rule you would actually run. Fourth, price every hour twice, with and without the candidate, and sum the year. The difference is that design's annual value on today's tariff, built from real hours rather than averages.

Fifth, and this is the step that separates analysis from advocacy, rerun the year against a stressed tariff: windows shifted later, the midday rate lower, the evening premium compressed, in the direction the grid's own evolution points. A design whose value survives the stress is robust; a design whose value collapses was a bet on a rate schedule, and the model should say so in exactly those words. The whole exercise fits in a spreadsheet a diligent analyst builds in a week, and its output is the honest sentence every sizing decision deserves: this design earns roughly this much on today's tariff, roughly that much on a plausible tomorrow's, and here is the hour-by-hour evidence.

One practical footnote from running this method on real sites: the binding constraint for non-export projects usually appears at the quietest hour, not the loudest. The overnight or weekend minimum, visible only in interval data, caps how much flat generation the site can absorb, and designs sized to impressive daytime peaks strand capacity at three in the morning. The 8,760-hour method catches this automatically, which is one more reason averages, which cannot, keep selling oversized machines.

Section 05Five sizing disciplines the shape imposes

• Model hours, not averages. One blended avoided rate hides the decision. 8,760-hour arithmetic against the actual tariff is the minimum bar.
• Name the driver. If the case is time-to-power or reliability, say so and let energy value be secondary; if the case is energy arbitrage, prove it against today's windows and a revised-tariff sensitivity.
• Respect the moving target. Tariff windows migrate and the grid fleet evolves. Prefer designs that stay sensible across a band of futures over designs optimal in exactly one.
• Check the minimum-load hours. For non-export projects, the cheap midday hours are often also your sizing constraint; oversizing to chase evening value strands capacity at noon.
• Keep the comparison neutral. The duck flatters storage in one conversation and firm generation in another. A study that only shows the hours favorable to its product has answered its own question.