Units, Percentages, And Denominators

A report says, “Our utilization is ninety percent, and our margins are improving.” That sounds encouraging. But ninety percent of what? Time allocated to a machine, time actually active, peak computation achieved, or time sold to customers? And what changed in revenue and costs? A precise-looking percentage can leave the important questions unanswered.

Before connecting the two claims, we need to understand what each number measures. This lesson is about units, percentages, and denominators. We'll first attach a unit and a measurement boundary to a number. Then we'll work through percentages whose meanings depend on the whole being counted. Finally, we'll reconcile storage conventions and

calculate a simplified energy charge without turning it into a complete operating budget. By the end, you should be able to reconstruct these calculations, spot double counting, and state what evidence a numerical claim still needs. Start with a unit: what kind of quantity is being described? Ten megawatts and ten megawatt-hours share a

familiar word, but they measure different things. Megawatts describe a power rate. Megawatt-hours describe accumulated energy. A facility's power draw tells us the rate at which it is using energy. To calculate how much energy it uses over an interval, we also need time. The number ten alone doesn't tell

us which of these questions was answered. The same distinction appears in model-related quantities. Ten thousand tokens is a quantity of units. Ten thousand tokens per second is a rate: a quantity divided by time. A count doesn't tell us how quickly it was produced, and a rate doesn't by itself tell us how much work occurred over an entire period.

Preserve the words and units attached to the number. They carry information that disappears if we compare two bare ten-thousands as though they meant the same thing. Ten dollars per successful task has another denominator. It describes cost divided by a count of particular outcomes. The word successful matters. Cost per attempted task, per generated response, and per successful task can describe different measures.

To interpret the last one, ask which tasks count as successful and which costs and time period are included. We will examine completion and cost more fully later. Here, the prerequisite is simply to notice what the division is actually counting. A denominator is the quantity by which another quantity is divided.

In a fraction, it is the quantity below the dividing line. The numerator is the quantity above it: what is being divided. To express a part of a whole as a percentage, divide the part by the whole and multiply by one hundred. For example, a ratio of

zero point six is sixty percent. For a percentage of time sold, we need both the sold time and the whole against which it is compared. Calling a figure “utilization” doesn't supply those definitions. Writing a percent sign after the answer doesn't supply them either.

Utilization can refer to several different measures. It might mean the fraction of time hardware is allocated, the fraction of time it is active, the fraction of its peak computation achieved, or the fraction of time sold to customers. These are not interchangeable questions. Something can be allocated without doing useful work throughout that allocation.

A measure of computation achieved is also different from a measure of time sold. Identify the definition, time period, and equipment or service boundary before comparing two utilization percentages. Availability concerns whether a service or component is usable under a specified definition. Consider this hypothetical machine. It is available for ninety-five percent of scheduled time.

A separate sixty percent of that available time is sold. The second percentage uses available time as its whole, not all scheduled time. Keep those two groups distinct. One tells us how much scheduled time was usable; the other tells us how much of the usable portion was sold. Pause and calculate.

The machine is available for ninety-five percent of scheduled time, and sixty percent of that available time is sold. What percentage of scheduled time is sold? You may find it easier to imagine one hundred scheduled hours. Work out the available hours first, then the sold part of those hours. Ninety-five percent of one hundred scheduled

hours is ninety-five available hours. Sixty percent of those ninety-five hours is fifty-seven hours. Equivalently, zero point nine five multiplied by zero point six equals zero point five seven, or fifty-seven percent of scheduled time. The other thirty-eight available hours were not sold, and five scheduled hours were unavailable. Those groups add back to the original hundred hours.

We have kept the denominator visible at every step. This multiplication is justified by the definitions in the example. We first select the available portion of scheduled time, then the sold portion within that available group. It isn't a rule to multiply any two percentages that appear in a report. If a different report already expresses sold time as a percentage of scheduled time, applying the

availability percentage again would answer a different question. You must establish what each percentage is a fraction of before combining them. That is also how double counting can enter a model. Suppose someone proposes an additional utilization reduction to account for the same unavailable time already removed by the ninety-five percent availability figure. The lost time would be removed twice.

Giving the second reduction a different label doesn't make it a different loss. An additional factor needs a genuinely distinct definition and boundary. Otherwise, the calculation can become falsely pessimistic by counting the same reduction again. A growth percentage needs an equally clear baseline. Suppose cost falls from ten dollars to eight dollars.

The reduction is two dollars. To express the reduction relative to the starting cost, divide two by ten. That gives zero point two, or twenty percent. The denominator is the original ten dollars. We aren't dividing by the final eight, because this question asks how much cost fell relative to where it started.

Now reverse the change. Cost rises from eight dollars to ten dollars. The increase is still two dollars, but the starting amount is now eight. Two divided by eight is zero point two five, or twenty-five percent. A twenty percent fall and a twenty-five percent rise can therefore describe the outward

and return journeys between the same two prices. The dollar change has the same size; the denominator has changed. Equal numerical changes need not have equal percentage changes. A percentage point measures the difference between two percentages. A relative percentage change instead compares the change with a starting value. These are two different descriptions, both potentially useful.

To keep them separate, first ask for the difference between the rates. Then ask how large that difference is relative to the original rate. Don't silently switch questions halfway through the calculation. Here is a second checkpoint. A conversion rate rises from ten percent to twelve percent. How many percentage points has it increased?

What is the relative percentage increase? Pause and name both answers, including the unit or description attached to each one. Twelve percent minus ten percent is an increase of two percentage points. The relative increase is two divided by the original ten, which is twenty percent. Think of the two rates expressed per

hundred opportunities: ten conversions versus twelve. The increase of two is one fifth of the original ten. This is a way to interpret the rates, not evidence that a particular experiment used one hundred people or established why conversion changed. Saying conversion improved “by two percent” would leave the intended meaning unclear.

Did the speaker mean a two-percentage-point increase, or a relative increase of two percent? Our example has a two-percentage-point increase and a twenty percent relative increase. Preserve that distinction when evaluating experiments. A correct subtraction can still be reported misleadingly if its result is given the wrong description. Ask what the reported change is measured against.

Storage conventions create another problem that can look like a disagreement between systems. A decimal gigabyte means one billion bytes. A gibibyte, the binary unit, means one billion, seventy-three million, seven hundred forty-one thousand, eight hundred twenty-four bytes. The exact counts are shown on screen. The symbols are G B for gigabytes

and G i B for gibibytes. A gibibyte contains more bytes than a decimal gigabyte. These are different conventions for describing amounts; the different label doesn't create or remove physical memory. A manufacturer's specification, a software display, and a spreadsheet may use different conventions. Reconcile them before deciding whether memory fits. Identify which convention each number uses,

translate the quantities to the same basis, and compare like with like. Don't infer missing capacity merely because two displayed numbers differ. Equally, don't assume matching-looking numbers describe identical byte counts. Unit reconciliation is a necessary part of interpreting the claim, rather than permission to ignore the actual requirement.

Now let's calculate energy from a power rate. Energy equals power multiplied by time. The units tell the same story: megawatts multiplied by hours gives megawatt-hours. If power draw varies, the amount of energy depends on that varying draw over the interval. This example avoids that complication by specifying a constant draw.

We should retain that assumption rather than treating a simplified example as a measurement of every facility's real behavior. In this hypothetical example, a facility draws twenty megawatts continuously for five hours. Electricity costs sixty dollars per megawatt-hour in the simplified calculation. We want two quantities:

the accumulated energy during those hours, and the energy charge at that stated price. The power rate, duration, and price per unit of energy are different inputs. Keep their units attached as you work. Pause for the energy checkpoint. At a constant twenty megawatts for five hours, how many megawatt-hours are used?

At sixty dollars per megawatt-hour, what is the simplified charge for that energy? Calculate the energy first, then apply the price. Leave the rest of the facility's operating budget outside this calculation. Twenty megawatts multiplied by five hours equals one hundred megawatt-hours. One hundred megawatt-hours multiplied by sixty dollars

per megawatt-hour equals six thousand dollars. The megawatt-hour unit in the energy quantity cancels with the per-megawatt-hour part of the price, leaving dollars. We used a power rate to find energy, then an energy price to find a charge. Multiplying twenty by sixty without accounting for the five hours would leave out a required part of the calculation.

Six thousand dollars is the simplified energy charge under these assumptions. It excludes demand charges, network fees, taxes, and other tariff elements. Those omissions are part of the interpretation, not a minor footnote that can be forgotten after the arithmetic. The example hasn't priced those items or established

which tariff applies to a real facility. To estimate an actual bill, the relevant charging terms and boundaries would need to be examined. It also isn't the facility's entire operating budget. The calculation addresses the specified energy quantity at the stated energy price. A complete operating-cost claim would need its own scope and evidence.

Naming an omission prevents a useful educational calculation from being presented as a broader answer than it supports. Correct arithmetic can coexist with an incomplete budget. The remedy is to state the boundary and investigate the missing terms, rather than changing a multiplication that was valid for its limited question. You now have a practical

way to read a technical number. Preserve the unit, identify the measured component or service, and state the period. For a rate, ask what quantity is being divided by what interval. For a percentage, identify the counted part and the whole. For a cost, identify both the included costs and the denominator.

For a calculation, list which assumptions connect the inputs and which items were excluded. This makes comparisons checkable instead of merely precise-looking. Return to our opening report: utilization is ninety percent, and margins are improving. We still cannot establish a relationship from the utilization percentage alone. Utilization might describe allocation,

activity, computation, or time sold. The commercial result also involves revenue and costs, including prices and contract terms. A high reported percentage is not, by itself, an explanation of improved margins. We need evidence about what changed within the relevant boundaries. Use this final checkpoint. Someone reports “our utilization is ninety

percent” alongside “our margins are improving.” What must you ask before linking those statements? Pause and organize your answer into measurement questions and commercial questions. Include useful output rather than stopping at the percentage. First define utilization: its time period, measurement boundary, numerator, and denominator. Identify whether it measures allocation, activity,

peak computation achieved, time sold, or another explicitly defined measure. Then inspect revenue, costs, prices, contract terms, and useful output. Ask what changed, over which comparable period, and what evidence connects those changes. The relationship cannot be established from ninety percent alone. These questions locate the missing information;

they don't assume that utilization can never matter. When making your own comparison, write down the unit and denominator before the conclusion. If you combine percentages, show that they refer to compatible groups and don't count the same loss twice. If you compare growth, preserve the starting value. If you compare storage, reconcile the byte convention. If you calculate a charge,

state its assumptions and omissions. These small acts keep the reasoning visible, so another person can check both the arithmetic and the question it actually answers. We have turned several impressive-looking numbers into defined, limited claims. Fifty-seven percent of scheduled time sold follows from the particular availability and sales definitions.

Percentage points and relative growth answer different questions.