When to Re-Run
the Study

A study answers a question on a date. The site is real, the tariff is the one on file, the fuel curve is the one published that month, and the recommendation follows from all of them as they stood the day the work was done. None of those inputs holds still, and the gap between deciding and energizing a large load is usually measured in years rather than months.

Owners tend to handle this in one of two ways, and both are mistakes. The first is to treat the study as settled. The document goes into a board pack, the recommendation becomes the plan, and eighteen months later capital moves against a set of assumptions nobody has looked at since. The second is the opposite reflex: every headline about grid conditions, fuel prices, or equipment costs reopens the whole analysis, the decision never closes, and the delay itself quietly becomes the most expensive line in the project.

Neither is a policy. A policy states, in advance, which inputs are allowed to change the answer, how far they have to move before they do, and what happens when they cross that line. This paper describes how to build one. The evidence below is drawn from federal, state, and utility publications current as of the access dates listed at the end. The cadences travel. The specific numbers do not, and an owner outside California should pull the equivalent documents for their own territory.

Section 01The revision that should end calendar thinking

The North American Electric Reliability Corporation publishes a Long-Term Reliability Assessment each year. It is the closest thing the continent has to a consensus ten-year view of electricity demand and resource adequacy, and it is assembled from the planning submissions of the regions themselves. It is not a fringe document, and it is not written to be provocative.

The 2025 assessment forecasts summer peak demand growth of 224 gigawatts over ten years, more than 69 percent above the 132 gigawatts projected one year earlier. Winter peak growth is forecast at 246 gigawatts. The compound annual growth rates for both are the highest since NERC began tracking them in 1995.¹ Most of the increase is attributed to new large-load additions, data centers foremost among them.

Read that as a statement about forecasting rather than about demand. A serious, well-resourced, industry-wide process looked at the same ten-year window twice, twelve months apart, and moved its central number by two-thirds. Any owner whose energy decision rests on an assumption about grid conditions three or four years out is resting it on the same class of estimate, produced with less information and less scrutiny.

The honest qualification matters as much as the finding. A continental forecast revision is not a fact about a particular site. A 69 percent change in aggregate demand growth does not mean a specific plant's answer changed by anything at all, and the utility circuit serving it may be entirely unaffected. Aggregate drift is a reason to check, never a reason to conclude. Distinguishing the two is the whole discipline, and the rest of this paper is about how.

Section 02A study is a dated instrument

Everything inside a study belongs to one of three durability classes, and separating them is the first practical step in any revalidation policy.

Findings that rarely move. The physical and legal facts of the site: who owns it, who holds the meter, how much land and clearance exist, whether a thermal host is present, what the lease term is, what sits adjacent. These are the findings that make a study worth commissioning once rather than annually. They change on transactions, not on markets.

Findings that move on a published schedule. Tariffs, incentive program steps, agency forecasts, and utility performance reports all have release calendars. They are knowable in advance, and a revalidation policy should be built on top of them rather than discovering them by accident.

Findings that move without warning. Equipment quotations, contractor availability, corporate capital allocation, and the owner's own load plan. In practice the last of these is the most volatile input in the entire model. Markets get the blame for stale studies. Internal forecast changes cause more of them.

Section 03Five clocks, not one

Owners often ask how long a study stays good, expecting a single number. There is no single number, because the inputs run on at least five separate clocks. The table below sets out the cadences that govern a California decision, with the source that publishes each.

Clock	Input	Where it is published	What a change can move
Nightly	Incentive program step levels	The Self-Generation Incentive Program publishes an incentive step tracker that is, in the program's own words, "updated nightly, or in the case of a lottery, after the results are published."2	Capital offset on storage and qualifying generation. Steps decline as budget is claimed, so this input can move with no calendar event at all.
Monthly	Near-term fuel prices	The federal Short-Term Energy Outlook, released monthly.3	Operating cost for every fuel-burning path, and the spread against grid-supplied energy.
Quarterly or on filing	Utility rates	Utility advice letters and rate change notices, filed as needed rather than on a fixed calendar.4	The value of everything measured against the bill: avoided energy, avoided demand, standby charges, export treatment.
Semiannual	Utility energization performance	Biannual energization reports required of the large California electric utilities under the Commission's 2024 energization decision.5	The credibility of the utility date the whole schedule case rests on.
Annual	Long-range demand and price outlooks; interconnection application windows	The Annual Energy Outlook, released April 8, 2026 for the current edition;6 the Long-Term Reliability Assessment;1 cluster application windows that open once a year.	The context the decision sits in, and in the case of application windows, whether the next opportunity is weeks or a year away.
Multi-year	Rate case outcomes and statute	General rate case decisions and enacted legislation.	Structural changes to rate design and incentive availability. Rare, and consequential when they land.

Two of those rows deserve elaboration, because they are the ones owners most often assume are stable.

On fuel: the August 11, 2026 Short-Term Energy Outlook expects the Henry Hub spot price to average $2.87 per million British thermal units in the third quarter of 2026, "down 50 cents compared with last month's forecast."³ That is a revision of roughly 15 percent to a near-term input in one monthly cycle, driven by production and inventory conditions the same publication describes as putting inventories at a record 3,985 billion cubic feet at the end of October.³ A study whose fuel-cost case was built on the July view was already carrying a materially different number by the second week of August.

On rates: a utility's own investor communications are a useful record of how often the bill changes. In a March 2, 2026 release, Pacific Gas and Electric Company described a reduction in residential electric rates effective March 1, 2026 as the fifth electric rate drop since January 2024.⁴ Those figures describe residential bundled service, and commercial and industrial rates move on their own schedules under their own filings, so the percentages should not be borrowed. The cadence should be. Five changes in roughly two years is the shape of the input, and any model that treats a rate as fixed for the life of an asset has made an assumption its author probably did not intend to make.

A study that cannot tell you what would change its mind has not finished its work.

Section 04What flips an answer, and what only moves a number

Most input drift is noise. Fuel moving 15 percent changes the operating cost line. It rarely changes which path an owner should take, because the alternatives moved with it or because the decision was never close on that axis. The distinction that matters is between inputs that move the number and inputs that move the ranking.

The way to know the difference is to compute it once, at the end of the original study, rather than to re-argue it every time something in the news moves. For each material input, the study should state the switching value: the level at which the recommended path stops being the recommended path. Gas at what price makes self-generation lose to grid service. A utility date slipping by how many months makes a bridge worth building. An incentive step closing at what point changes the storage case. Those numbers are cheap to produce while the model is open and expensive to reconstruct later.

A study delivered with its switching values stated converts revalidation from a judgment call into an observation. Nobody has to decide whether a headline is significant. The published number either crossed a stated line or it did not. This is the same discipline as labelling every figure firm or illustrative. It moves the hard thinking to the moment when the analyst has the full context, and leaves the owner with a test they can run in an afternoon.

It also exposes the studies that cannot survive it. A recommendation with no stated breakpoints is either extremely robust or was never tested, and from the outside those look identical.

Section 05Drift is not symmetric across technologies

A revalidation policy has to cover every path that was priced, not only the one that won. Each carries a different exposure to the clocks above, and the honest case for and against each is largely a case about which inputs it is hostage to.

Continued utility service. For: it is insulated from fuel markets, equipment quotations, and supply chains entirely. The only inputs that matter are the utility's date and the rate schedule, which makes revalidation cheap and narrow. Against: it is the path most exposed to schedule drift, and schedule is the input the owner has the least ability to influence. Over a long horizon it is also fully exposed to rate design changes, which arrive through filings the owner does not control and often does not see.

Storage. For: capital cost trends and incentive programs have both worked in owners' favour in recent years, so a decision revisited later has sometimes improved rather than decayed. Against: it is the path most exposed to incentive-step risk, because program steps deplete on subscription rather than on a calendar. An owner waiting for a better answer can lose a step while waiting. Storage also inherits the reliability of whatever service it firms, and revalidating the battery without revalidating the underlying service is a half-done job.

Solar. For: there is no fuel line to revalidate at all, and the generation profile of a given site is about as stable an input as this field offers. Against: its value is almost entirely a function of tariff structure and time-of-use period definitions, which sit on one of the faster clocks. A tariff revision can move a solar case substantially without a single physical fact changing.

Reciprocating engines. For: comparatively short lead times mean a decision can be made later with a smaller penalty, which is itself a hedge against uncertainty. Multi-fuel capability, where it exists, hedges the fuel line directly. Against: this is the path most exposed to the monthly fuel clock and to air district rule revisions, and those two move independently of each other and of everything else in the model.

Gas turbines and microturbines. For: where a genuine thermal host exists, the heat recovery credit is one of the more stable value lines in an energy model, because it is tied to a physical process rather than to a market. Against: the same fuel exposure as engines, with efficiency characteristics that can make the sensitivity to a gas price revision steeper rather than shallower.

Fuel cells. For: higher electrical efficiency dampens sensitivity to fuel price revisions per unit of output, and the permitting posture is frequently the least contested of the combustion-adjacent options. Against: capital intensity is high, and the commercial structure typically involves long-term service pricing whose assumptions run well past any horizon that can honestly be revalidated. More of the decision therefore rests on inputs that no future check will resolve.

Linear generators and other newer platforms. For: this is the category where published cost and performance data improve fastest, so revalidation genuinely can turn a marginal case into a good one. Against: a thin operating base means the figures being revalidated against are themselves less settled, and a favourable revision carries less evidentiary weight than the same revision would for a mature platform.

Because the exposures differ, revalidation has to be run across the full option set. Refreshing the inputs only for the recommended path, and comparing the result against stale numbers for the alternatives, is how an independent study quietly turns into a document that defends a prior conclusion.

Section 06The trigger list

These are the events that should prompt a revalidation. A policy that names them in advance is worth more than one that relies on somebody noticing.

1. The load plan changed beyond the study's stated tolerance.The most common trigger and the most often missed, because it originates inside the organization rather than in the market. A change in the equipment list, the shift pattern, or the expansion timeline can invalidate a sizing case entirely.
2. The utility's written date moved.Any revision to a documented energization date, in either direction. A date that improves is as much a trigger as one that slips, because it can remove the need for a bridge.
3. The tariff the model was built on was superseded.New rate schedules, revised time-of-use periods, or changes to standby and departing-load provisions. The model cites a specific schedule, and when that schedule is replaced the citation is stale by definition.
4. An incentive step closed or a statutory provision changed.Program steps deplete without notice. Statutory changes are rarer and larger. The federal investment tax credit stands at a flat 30 percent for qualifying property under current law as of August 2026, with no adders assumed in any figure here. That is a statement with a date on it, and it should be confirmed with tax counsel rather than carried forward.⁷
5. A stated switching value was crossed.The breakpoints from Section 04, checked against published data. This is the trigger that requires no judgment, which is exactly why it is worth defining in advance.
6. The site changed.Ownership, lease term, meter configuration, adjacent development, or anything affecting site control. These reach the durable findings, and a change here can invalidate more of the study than any market movement.
7. A permit or interconnection assumption was tested and failed.The first real agency response is worth more than any assumption that preceded it. It should be treated as new evidence immediately, not filed until the next review.
8. Twelve months elapsed with no other trigger.A backstop, not a schedule. If a full year passes and nothing on this list has fired, the inputs should still be refreshed once before capital moves, because the absence of a noticed trigger is not evidence that nothing moved.

Section 07Three tiers of re-run

Revalidation is not a synonym for a new study, and treating it as one is how organizations end up paying repeatedly for work they already own. There are three distinct tiers, and matching the tier to the trigger is most of the cost control.

An input refresh updates the published values and re-runs the existing model. Fuel curves, tariff rates, incentive steps, and the current utility date go in. The structure of the analysis does not change. If the original study carried a source and an access date against every load-bearing number, this is a mechanical exercise measured in hours. If it did not, the same work becomes archaeology, and the cost of that archaeology is the real argument for sourcing discipline in the first place.

A sensitivity re-run is warranted when a switching value has been crossed or a material assumption has been contradicted. The option set is unchanged, but the ranking is genuinely back in question, and the comparison has to be rebuilt across all paths rather than adjusted on one.

A full re-study is warranted only when the question itself changed. A different load, a different site, a different corporate objective, or a constraint nobody knew about at the outset. This is rare, and an advisor who proposes it in response to an ordinary market movement should be asked which part of the original question is no longer the question.

Section 08When not to re-run

Decision churn has a price, and it is usually paid by the schedule. Every reopened analysis pushes the energization date, and in a market where the queue is the binding constraint, months surrendered to further study are rarely recovered later. Waiting for a better number is a decision, and it should be priced against the no-project baseline like any other.

There is an organizational cost as well. A board that receives a reversed recommendation twice in a year begins, reasonably, to discount the analysis rather than the conditions. Credibility spent on avoidable revisions is not available later for the revision that actually matters.

The practical guard is to refuse three specific re-runs. A re-run prompted by a headline rather than by a published figure crossing a stated line. A re-run of the recommended path alone, without refreshing the alternatives on the same basis. And a re-run requested by a party who benefits from the delay or from the conclusion changing. None of those is analysis. Each is a decision being made by something other than evidence.

The cleanest version of all of this is set at delivery, not discovered later. A study should state its own expiry conditions on the day it is handed over: which inputs it depends on, where each is published, how often each is revised, and what value each would have to reach to change the recommendation. An owner holding that document does not have to wonder whether the analysis is still good. They can check, in an afternoon, against public sources, and get an answer with a date on it.