Anatomy of a Power Readiness Study:
What a Bounded Decision Should Contain

Most executives have read a hundred building inspections and exactly zero power studies, so the first one that lands on their desk gets judged by weight and confidence rather than anatomy. This paper supplies the anatomy: the six components a bounded power readiness study must contain before anyone acts on it, and the honest boundary of what two weeks of disciplined work can and cannot resolve.

Section 01The product is a decision, not a report

A power readiness study exists to answer one narrow, consequential question: which path to power should this specific site take, by what date, under what conditions, and what would change the answer. Every page in the document either serves that question or pads it.

Bounded means three things are fixed before work begins: the question, the fee, and the window. In our practice the window is about two weeks. The bound is not a discount mechanism; it is a materiality discipline. An open-ended study drifts, because every open question invites another, and unbounded scope is always paid for twice, once in fees and once in months. A bounded study is forced to rank its questions by a single criterion: whether the answer would move the decision. Questions that fail that test are recorded as open items for the next phase, not answered at leisure inside this one.

It is equally important to say what the study is not. It is not an interconnection application, and it holds no place in any utility queue. It is not detailed engineering, not a permit filing, and not a tax opinion. It is the document that decides whether those far larger expenditures happen at all, in what order, and aimed at which path.

The anatomy has six components, in a fixed order: a measured load basis; a complete paths matrix, including the paths that pay the analyst nothing; sensitivity analysis on the assumptions that actually move the answer; a permitting and fuel screen built from site facts; an incentive screen at statutory values; and a decision memo that names the conditions under which its own recommendation fails. Each component has a minimum standard, and each has a characteristic defect that exposes a defective study in minutes. The rest of this paper takes them in order.

Section 02Component one: a measured load basis

Every number downstream of the load inherits the load's errors, which is why the load basis comes first and why it must come from the meter rather than from a list of equipment ratings. In Pacific Gas and Electric Company territory, up to 48 months of historical interval usage is available through the utility's data-sharing platform, generally at fifteen-minute granularity for commercial accounts.¹ Authorizing that access is a first-day task. There is no analytical substitute for four years of the site's actual behavior in fifteen-minute slices.

What the interval record must establish is shape, not just volume. At minimum: a full year of coverage, because seasonality decides sizing; the demand profile rather than energy totals alone, because two sites with identical annual consumption can carry very different peaks; the true coincident peak, which is almost always well below the sum of equipment ratings; and the overnight and weekend floor, which determines how many hours of the year a continuously running machine would actually be covered by on-site demand. In practice the floor decides more sizing questions than the peak does.

New facilities and expansions have no meter history, and the honest response is a labeled proxy, not false precision. A defensible proxy states its provenance: equipment schedules, design documents, commissioning records from a comparable facility. It is then treated as an assumption and stressed in the sensitivity work, because planned loads routinely arrive later, and ramp slower, than their design basis.

The characteristic defect is nameplate arithmetic: connected equipment summed, multiplied by a diversity guess, and presented as the load. Everything downstream of that number oversizes, and oversizing flatters every capital-heavy answer on the menu. An assumed load is the most common way a study smuggles its conclusion into its inputs.

Section 03Component two: the paths matrix, including the rows that pay nothing

The matrix is the center of the document: every credible path to power for this site, priced side by side from sourced and dated inputs. The menu is longer and less standard than it has ever been. California's utilities regulator approved an interim framework in July 2025 to streamline connections for large new loads, and utility service offerings now trade firmness for speed in ways that did not exist a few years ago.² In that environment the decision lives in the matrix, not in any technology chapter.

Paper No. 001 in this series sets out the full menu and the test that goes with it; the recital here is brief. Eight families of answer recur: full utility service, flexible or curtailable service, phased energization, storage as a bridge, on-site generation, hybrid portfolios, relocation, and no project. A real site rarely sustains all eight as live candidates, but each one leaves the matrix only in writing, with a dated reason. Written eliminations are what make the study auditable a year later, when a fact has changed and the owner needs to know whether the change reopens the decision.

Three of those rows deserve their own sentence, because they are the rows that pay the analyst nothing: plain utility service on the documented timeline, relocation of the load to where power already exists, and the deliberate no-project case with the cost of waiting priced. No equipment seller can afford to present them well. An independent study must, because they are the benchmark every powered path has to beat. Their absence is the fastest tell that a document is a proposal wearing a study's clothing.

Inside the on-site generation family, neutrality has to survive contact with real equipment classes, and the discipline is symmetrical: the study argues against every candidate it prices, including the eventual winner. Reciprocating engines are typically the cheapest capital per unit of capacity with the deepest service ecosystem, and they are combustion machines whose permits in strict districts carry real control-technology burdens. Combustion turbines and microturbines are compact and yield high-grade heat, and their efficiency at small scale and part load deserves scrutiny. Fuel cells run quietly at high electrical efficiency with near-zero criteria-pollutant emissions, and they carry higher capital cost, gas-price exposure, and a stack-replacement lifecycle that belongs in the model. Linear generators are modular and fuel-flexible, and their shorter fleet history shifts the diligence burden onto operating data. Solar with storage buys no fuel at all, and its capacity factor means it rarely carries a continuous load alone. Each of those sentences cuts both ways on purpose. A matrix that only argues for things is not an analysis.

Section 04Component three: sensitivity on the assumptions that move the answer

A matrix of point estimates is a photograph of one possible future, and the future has a habit of leaving the frame. Sensitivity analysis is where a study confesses what it cannot know, and it is the component most often missing in the wild.

The national benchmark price of natural gas averaged $2.03 per million Btu in 2020, $6.45 in 2022, and $2.19 in 2024.³ Any fuel-consuming path priced at a single gas price in any of those years embedded a threefold error band without disclosing it. The remedy is not a better forecast; nobody has one. The remedy is to price fuel-burning paths across a defensible range of delivered prices at the meter and then publish the crossover: the delivered price above which the self-supply path loses to the best alternative that burns nothing. That single number converts an argument about forecasts into a monitoring instruction.

Three assumptions move the answer in most studies of this kind. Delivered fuel price is the first. The energization date is the second: the grid path is priced at the utility's documented date, and priced again with that date slipped by one year and by two, because congested pipelines move dates and a plan that dies on a twelve-month slip is not a plan. Load realization is the third: the ramp actually achieved against the ramp assumed. Other inputs matter at the margin; these three flip recommendations.

The method is unglamorous. Move one assumption at a time across its range and record which movements flip the ranking; then combine the two worst credible movements and see whether the recommendation survives. What matters most is where the flips are reported. The page-one rule: if the answer changes anywhere inside a plausible range, that condition belongs in the decision memo itself, stated as a trigger, not buried in an appendix exhibit. A flip condition is not a weakness of a study. Concealing it is.

Illustratively: if a generation path beats its grid alternative at a delivered gas price of eight dollars per million Btu but loses at eleven, and the plausible delivered range spans both numbers, then the honest recommendation is conditional, and the memo names the published index the owner watches and the level at which the decision reopens. Owners are well served by conditional answers with named triggers, and badly served by unconditional answers built on quiet assumptions.

Section 05Component four: the permitting and fuel screen

A path that cannot be permitted or fueled at the site is not a path; it is a brochure page. The screen's job is to convert generic candidates into site-specific ones, using named rules that are current as of a stated date.

Air permitting is the sharpest filter in California, and it is district-specific. In the Bay Area district, for example, new source review triggers a best-available-control-technology determination for any new or modified source with the potential to emit ten pounds or more per day of specified criteria pollutants,⁴ a threshold that meaningful combustion installations generally engage. Non-combustion electrochemical equipment presents a lighter criteria-pollutant profile under rules of that shape, though it still faces building, fire, and interconnection reviews of its own, and a lighter permit does not offset capital cost by itself. Neither fact settles the technology choice; the screen prices and schedules it. The same machine can face materially different treatment across a district line, which is why a screen that would read identically in any county was not written about your county.

The fuel screen is the air screen's twin. Gas adequacy at the meter is a fact to confirm with the gas utility, not to assume: delivery pressure and volume at the point of use, and the lead time and cost of any upgrade if the existing service falls short. Liquid fuels bring storage rules and fire code. Solar brings land, roof structure, and shading reality. Every candidate brings noise ordinances, and some bring water. And any equipment that operates in parallel with the grid takes an interconnection review of its own even though it serves load behind the meter; the screen states which application applies and what its posture tends to be, labeled as an estimate.

The deliverable is one paragraph per surviving path: the permit class it lands in, the rules that govern it by name and date, a timeline expectation stated as a range and labeled as an estimate, and the single largest permitting risk. That is not a permit application. It is the map that says where applications go next and which one sits on the critical path.

Section 06Component five: the incentive screen at statutory values

Two rules govern this component: statutory values only, and a date on every line.

Under current federal law, the investment tax credit for qualifying clean-energy property stands at 30 percent (26 U.S.C. §§ 48, 48E).⁵ Qualification is conditional and fact-specific: the statute attaches requirements to that rate, and it provides bonus adders that exist on paper for particular circumstances and must be individually qualified for a specific project before they are worth anything. A disciplined screen books the statutory case for paths whose property plausibly qualifies, lists each adder as unqualified upside alongside the condition it depends on, and directs the qualification itself to qualified tax counsel. The study screens; counsel qualifies. A study that does counsel's job has exceeded its competence, and a study that books adders it has not qualified is manufacturing a winner.

The characteristic defect is adder-stacking: a path that wins only when every hoped-for bonus lands. If the ranking reverses at statutory values, the study reports the path as losing, with the adders as documented upside that counsel may later confirm. The base case is where decisions are made; upside is where financings improve.

Program-level incentives deserve a sentence of humility. State and utility programs that once funded on-site equipment in California have stepped down, closed to new applications, or changed shape repeatedly over the past two decades, which is why every incentive line carries an as-of date and why a screen older than its own study is stale by definition. For public-agency owners, procurement and financing routes are agency-specific and counsel-led; a study for a public site says that plainly and hands the question to counsel rather than paraphrasing it.

Section 07Component six: the decision memo

The final component is the shortest, the hardest to write, and the reason the owner paid. It contains three things: one recommended path; the conditions under which that recommendation changes, stated as named, monitorable triggers drawn from the sensitivity work; and the next three actions, each with an owner and a date.

A study that cannot name the conditions under which its own recommendation fails has not finished the job.

The memo's governing discipline is falsifiability. "Proceed on the hybrid path, provided the documented energization date holds beyond the deadline and delivered gas stays inside the modeled band; if the utility advances its date by a year, the grid path wins, and this memo says so now" is a decision. "Several attractive options merit further evaluation" is a refusal to decide, invoiced as if it were one.

Three actions, not thirty. Each next action should be the cheapest available test of the largest open assumption: request the utility's written capacity and timing statement; obtain the gas utility's confirmation of delivery at the required pressure and volume; authorize the interval-data pull for the comparable facility the load proxy leaned on. Cheap tests first is not timidity. It is how a two-week study keeps its conclusions honest a month after delivery, without a second engagement.

Section 08What two weeks resolves, and what it deliberately does not

Bounded work has a boundary, and stating it is part of the product. The split below is the honest one, and a study that blurs it in either direction is misdescribing itself: claiming outcomes it cannot deliver, or quietly expanding into billable work the decision never needed.

Two more things sit outside the boundary for a different reason: they are the owner's to move. The speed of the load basis is set by how quickly data access is authorized, and the strength of the utility rows is set by whether the site's service correspondence exists on paper rather than in the memory of a phone call. A two-week study is downstream of the owner's own files, and the best owners treat the intake checklist as seriously as the deliverable.

None of this anatomy is proprietary, and that is deliberate. A specification only protects the buyer if the buyer can check it. Bring this list to any power study, whoever writes it, including ours: six components, sourced and dated inputs, estimates labeled, flip conditions on page one, and a recommendation someone signed with their own name. Documents built that way survive their first contact with a board, a lender, and a regulator. Documents built otherwise are weight.