The Load Basis:
Why Interval Data Decides
More Than Technology

The most consequential input to an on-site power decision is not the technology comparison, the fuel forecast, or the incentive model. It is the description of the load itself, and at most California business sites that description already exists, recorded every fifteen minutes by a meter the owner has never queried.

Section 01Three descriptions of the same load

Every energy proposal that crosses an owner's desk is built on a load basis: some statement of what the site consumes, when, and how hard. Only three kinds of load basis exist in commercial practice, and they are not interchangeable. One describes what the site could draw. One describes what it did draw, with the clock removed. Only the third describes what the site actually does, hour by hour and season by season, and it is the only one of the three that can size a power plant.

The nameplate basis is the sum of the ratings on connected equipment: every motor, chiller, rectifier, and panel schedule added up. It is the right tool for the job it was designed for. Electrical infrastructure is sized on nameplate logic because wire, switchgear, and transformers must survive the worst simultaneous case, and conservatism there is a safety requirement, not a bias. The same conservatism becomes a defect the moment nameplate arithmetic is used to size generation. Real facilities never run everything at once. Engineers call the gap diversity, and it is routinely large: the measured peak of a working site typically lands well below the sum of its ratings, and the average load lands well below the measured peak. A generation plant sized to nameplate is a plant sized to a fiction, and the error compounds over the asset's life, because oversized machines run at part load, where combustion classes in particular give up efficiency and accumulate wear per useful kilowatt-hour.

The billing basis is the utility bill: twelve monthly energy totals, a demand figure, and a rate schedule. Bills are honest documents with real uses. They are the ledger a study must reconcile against, the fastest sanity check on any model, and a sufficient basis for budget-level screening. What a bill cannot do is locate energy in time. Two sites can post identical annual kilowatt-hours while one runs flat around the clock and the other spikes to triple its average every afternoon. Every downstream question that matters (which technology, what size, whether storage earns its place, how the tariff lands) turns on that difference, and the bill is silent on it.

The interval basis is a time series from the revenue meter itself. In Pacific Gas and Electric Company territory, the meter already records business electric usage in fifteen-minute intervals and residential usage hourly, and the customer's online account displays usage roughly one day after it occurs.^1,2 Ninety-six observations a day. More than thirty-five thousand a year. The shape of the operation is in there: the morning ramp, the shift structure, the weekend floor, the season, the one compressor that never shuts off.

That ratio is the information gap in most power decisions. A year of business interval data carries nearly three thousand times as many observations as a year of bills. Nothing else an owner can obtain before spending capital carries as much decision-relevant information at as little cost, and it is already being recorded.

Section 02Load factor is the screen; shape is the answer

The first number to compute from interval data is load factor, which the U.S. Energy Information Administration defines as "the ratio of the average load to peak load during a specified time interval."³ It is a plain division, and it is the fastest single-number description of how a site uses capacity. A load factor near 1.0 describes a site that continuously uses what it peaks at. A low load factor describes a site that buys capacity it rarely uses.

Load factor sorts the field before any seller enters the room. Continuous-process manufacturing, cold storage, water treatment, server floors, controlled-environment agriculture: operations like these can run load factors high enough that capacity purchased is capacity used nearly every hour, which is the condition under which always-running generation classes recover their capital sensibly. Single-shift plants, schools, offices, and event venues run low load factors: their peaks are short, their nights are quiet, and machinery sized to the peak would idle for most of its life. Neither profile is better. They are different problems, and they buy different solutions.

But load factor is a screen, not an answer, because it is one number and shape is a curve. Two sites at the same load factor can differ decisively. The tool that exposes the difference is the load duration curve: sort every interval of the year from highest to lowest and read the result as hours at or above each level. A flat curve says the site needs most of its capacity most of the time. A curve with a steep knee says the top slice of demand exists for only tens of hours a year, which is precisely the slice that storage, scheduling, or tariff management can address without any generation at all. Where the knee sits, and how many hours sit above it, is frequently the most valuable single fact in the entire study.

Interval data also answers the questions load factor cannot. The seasonal structure shows whether the peak is a July afternoon or a harvest campaign. The week structure shows whether the site sleeps on weekends. The overnight floor matters because an always-on machine must follow the trough, export, or curtail, and its turndown limit measured against that floor decides whether continuous operation is mechanically sensible at all. And the distinction the tariff monetizes, demand as a rate in kilowatts against energy as a quantity in kilowatt-hours, is recorded natively by the meter and priced separately on most commercial schedules.

Section 03Shape picks the technology, not the reverse

Once the shape is on the table, technology selection loses most of its mystery, because each generation and storage class has shapes it serves well and shapes it serves badly. The honest version of that mapping follows. Classes are named generically because the mapping belongs to physics and finance, not to any manufacturer, and a tendency in this table is a starting point for site-specific analysis, never a verdict.

Read the table in both directions. Every class in it is the right answer to some shape and the wrong answer to others, which is why a recommendation that arrives before the interval data is a tell. Sellers do not usually misrepresent their machines; they meet a shape argument coming the other way. The discipline is that the shape must come from the meter, not from the pitch.

Two sites with identical annual consumption can justify opposite power plants. The difference never appears on a bill, and it is unmistakable in interval data.

Section 04One annual total, two opposite answers

A single deliberately simplified comparison shows how much of the decision rides on shape. The figures that follow are illustrative, chosen for arithmetic clarity rather than drawn from any site.

Site A runs a continuous process around the clock at a load factor near 0.9; its demand moves inside a narrow band all year. Site B posts the same annual kilowatt-hours from a single-shift operation whose afternoon peak reaches roughly three times its average, a load factor near 0.35. On every bill, the two sites are twins.

For Site A, a generation class that prefers steady operation can be sized near the flat demand and run almost every hour of the year at or near its best point. Fixed costs spread across the maximum possible kilowatt-hours, the grid covers maintenance windows, and the machine's dislike of cycling never gets tested. This is the shape that always-running classes were built for, and the capital mathematics reflect it.

For Site B, the same machine faces a choice of failures. Sized to the peak, it runs at a utilization equivalent of roughly a third of its potential, so each delivered kilowatt-hour carries close to three times the capital burden it carried at Site A, before counting the part-load efficiency penalty that combustion classes pay and the cycling duty that electrochemical classes prefer to avoid. Sized to the base instead, it leaves the peak to be solved by something else, which is often the correct engineering answer and turns the project into a hybrid. And at Site B the stronger candidates may never have been generation-first at all: a storage system whose duration matches the afternoon peak, schedule and process adjustments, tariff work, and grid capacity for the remainder deserve to be priced first. The arithmetic is illustrative; the direction is general.

Identical bills, opposite plants. Nameplate could not have separated them, and annual totals did not. Only the meter knew.

Section 05Obtaining your own interval data in California

California put the customer in control of this data more than a decade ago. In July 2011 the California Public Utilities Commission adopted Decision 11-07-056, the privacy and security framework governing customer electricity-usage data at the state's three large investor-owned utilities, which protects that data and conditions third-party access on customer authorization.⁴ The practical machinery built on that framework is now routine, and an owner can exercise it without engaging anyone.

At Pacific Gas and Electric Company the mechanics are as follows, described here factually, as the utility documents them, because programs are the utility's to define and revise. The meter records business electric usage in fifteen-minute intervals and residential usage hourly, and an online account displays usage data roughly one day after it occurs.¹ The Share My Data platform then offers two routes. A customer can download their own usage history directly. Or the customer can authorize a registered third party to receive electric interval data, gas data, and billing information over a secure interface, with the authorization scoped by data category, by service identifier, and by duration, and cancellable by the customer at any time from the same online account.²

The pattern is not proprietary to one utility. It is an implementation of the national Green Button framework, built on the Energy Services Provider Interface standard: Download My Data for self-service files, Connect My Data for authorized, revocable, ongoing transfer to a third party.⁵ The state's other large investor-owned utilities operate under the same Commission privacy decision and maintain their own data-access channels; the mechanics differ by utility, and current details should be confirmed on each utility's own pages before a study relies on them.

What to request is simpler than the plumbing. Twelve months of interval data is the minimum on which a defensible study can stand; twenty-four months is better, because it contains two summers and a full cycle of anomalies to explain. Pull every service agreement and every meter serving the site, not only the largest. Pull billing history alongside the intervals, because the study must reconcile the time series against billed totals and the rate schedule before anything downstream is trusted.

Two boundary cases recur. First, authorization follows the account: a tenant behind a landlord's master meter is not the utility's customer for that meter and will need the account holder to authorize or supply the data. Second, the revenue meter sees only its own boundary: loads submetered inside the fence, and equipment not yet installed, are invisible to it, and the study must say so rather than treat the meter's record as the whole story.

When no meter history exists

New sites and new loads have no interval history, and the discipline then is a hierarchy of proxies, each labeled as what it is. The best proxy is interval data from a comparable facility the same owner already operates, adjusted for stated differences. Next is an engineering build-up from equipment schedules with explicit diversity assumptions and an operating calendar, presented as an estimate with its assumptions on the page. Public modeled profiles exist as a starting shape: the U.S. Department of Energy's End-Use Load Profiles project publishes hourly and sub-hourly electricity profiles for the major residential and commercial building types, calibrated so that model output matches usage observed by regional grids.⁶ A modeled profile is a reasonable place to begin and an indefensible place to stop. Whatever the proxy, the study's sensitivity section must show what happens to the recommendation if the assumed shape is wrong, because when a proxy sits in the load basis, shape risk is the project's largest unpriced risk.

Section 06Seven checks against an assumed load

Assumed loads fail in recognizable, repeating ways. The following checks catch the seven failures we see most, and they bind our own work as much as anyone's. Run them against any load basis before capital moves.

1. Reject the stack of nameplates.If "the load" is a sum of equipment ratings, diversity has been ignored and the plant is being sized to a worst case that never occurs. The symptom is proposed on-site capacity that dwarfs anything visible on the utility bill.
2. Reject the flat-load fiction.Annual kilowatt-hours divided by 8,760 hours is an average, not a profile. Treating it as constant demand flatters always-running generation, erases the peak that sizes equipment, and hides the overnight floor that decides turndown.
3. Distrust a single year.One year is one weather draw and one production draw. Check the sample year against production records, prefer twenty-four months, and explain anomalies instead of averaging them into the base case.
4. Refuse borrowed shapes.A profile imported from another industry, climate, or shift structure is a costume. Load shape is operational: two facilities of the same type and size can carry decisively different curves.
5. Interrogate the peak.Fifteen-minute averages smooth the instantaneous inrush that electrical design must still survive, while a one-time event can leave a demand spike that oversizes a machine for a decade. Pull the top fifty intervals and establish when, how long, and why.
6. Put growth in scenarios, not the base case.Sizing day one to a hoped-for expansion buys idle capital; ignoring a documented fit-out ramp buys an undersized plant. Growth enters as named scenarios the owner signs, with dates, never as a silent assumption.
7. Reconcile to the meter boundary.Decisions live at meters, not at "the site." Multiple service agreements, master meters, and submetered tenants must be mapped, and the load basis must tie back to what each meter records and bills.

Section 07The load basis chapter comes first

In a defensible study the load basis is a chapter with a fixed anatomy: measured intervals reconciled against twelve to twenty-four months of bills; the load factor, stated with its period; the load duration curve and where its knee sits; the seasonal and week-structure decomposition; the table of top peak intervals with their causes; the overnight floor and what it implies for turndown; growth as owner-signed scenarios; and every proxy labeled, with sensitivity showing what changes if the shape is wrong. None of this is exotic. All of it is checkable, which is the point.

It comes first in the document because everything downstream inherits it. The technology comparison, the sizing, the storage case, the tariff analysis, and the financial model are all functions of the shape; corrupt the input and the rest is confident nonsense. That is also where independence earns its keep. A seller can survive an assumed load, because the proposal is signed before the shape is tested. The owner operates the consequences for twenty years.

The practical instruction of this paper fits in one sentence. Before anyone proposes anything for your site, pull your own interval data: the meter has been keeping the record all along, the state's rules put it in your hands, and it is the least expensive decision-grade information you will ever hold.