Reading Vendor Models:
Behind-the-Meter Economics
Without Guarantees

A behind-the-meter pro forma is the most persuasive document in the energy business because it looks like the least persuasive kind: a spreadsheet. This paper is a field guide to the six places those spreadsheets flatter themselves, and a line-by-line checklist for the owner who has to carry the numbers to a board.

Section 01The document you are actually holding

Late in every behind-the-meter sales conversation, a model arrives. It is usually twenty to forty rows: a capacity, an annual output, a utility rate, an escalation percentage, a fuel or service line, an incentive block, and, in bold near the bottom, a payback year and an internal rate of return. Boards remember the bold row. Almost nobody interrogates the rows that produced it.

The first paper in this series made a structural observation: a seller's feasibility analysis is a sales document, whatever its cover page says. This paper extends that observation into the spreadsheet itself, because the mechanism deserves respect. It is not fraud, and it is rarely even conscious. Nearly every input in a generation or storage pro forma has a range a competent professional could defend. Availability, escalation, incentive value, avoided rates, residual value: each has a low end and a high end, and a seller's model selects from the favorable end of each range, one defensible choice at a time.

Six patterns account for most of the flattery we encounter in study work, across every technology family: availability and degradation treated as footnotes; asymmetric escalators; the quiet omission of standby and departing-load charges; incentive adders booked before they are qualified; savings computed against a blended average rate; and terminal value doing silent work in the final years. The sections that follow take them in order. None requires an engineering degree to catch. Each requires only the discipline to ask for a specific document, and to treat its absence as an answer.

No single cell of the model is indefensible. The compound of twenty favorable cells is a machine that cannot lose on paper.

Section 02Availability and degradation: the assumption that runs the model

Every pro forma begins, visibly or not, with an energy line: the kilowatt-hours the equipment produces each year. That line is nameplate capacity multiplied by hours, discounted by two factors the model may or may not disclose. Availability is the fraction of hours the machine is capable of running. Degradation is the decline in output and efficiency between overhauls and over the equipment's life. Small-looking differences matter because the energy line multiplies through every savings row beneath it. The gap between 98 percent and 92 percent availability is roughly three weeks of output a year; across a twenty-year term it is a material share of the modeled return. Both figures in that sentence are illustrative. The point is the sensitivity, not the specific spread.

The first question is which quantity the model actually used. Availability and capacity factor are different numbers: a machine can be available and not running, and a model that applies an availability figure where a capacity factor belongs has overstated the energy line before the pricing debate even begins. The second question is whose number it is. Fleet-average figures describe a fleet, under mature service programs, with established spares logistics. The model in front of you is about one site, one fuel or solar resource, and one maintenance calendar.

The honest profile differs by technology family, and each family has a real case on both sides:

• Reciprocating engines can post strong availability under a disciplined service program, and the service ecosystem is the deepest of any class. They also carry fixed minor and major overhaul intervals measured in running hours; a continuous-duty engine reaches those intervals quickly, and the model should show outage days and overhaul costs in the years they occur rather than amortizing them into invisibility.
• Gas turbines and microturbines run long intervals between major inspections at steady load. Output derates with ambient temperature and altitude, and efficiency falls at part load; a model built at rating-sheet conditions and constant full load describes a test cell, not a site.
• Fuel cells avoid combustion's mechanical outage drivers and run quietly at high electrical efficiency. Their electrochemical stacks degrade with operating hours: output and efficiency decline between stack replacements, and each replacement is a scheduled, priced event. A fuel-cell pro forma showing year-one output in year fifteen with no stack event in between has skipped a chapter.
• Linear generators are modular, and modularity genuinely helps: units can be serviced in rotation, which softens single-point outages. As a newer class, the public operating history is shorter, so the availability assumption deserves verification against contract terms rather than reputation, in either direction.
• Solar and storage carry no fuel and few moving parts, and forced-outage behavior is genuinely favorable. The constraint is the resource itself: production follows weather and season, panels degrade slowly but measurably over decades, and battery capacity fades with cycling until augmented. The honest model shows production year by year, declining, with augmentation priced. Resource variability is not an outage; it is the technology's shape, and it belongs in the energy line either way.

Sellers sometimes answer this section by pointing to a contractual availability guarantee. Read what the provision pays. Most such clauses remedy a shortfall with service credits or liquidated damages; the remedy is real, but it is a different asset from the kilowatt-hours the spreadsheet already monetized at the tariff rate. The ask here is simple: the maintenance calendar in hours, the degradation curve in writing, and an annual energy line that reconciles to both.

Section 03Escalator asymmetry: compounding does the selling

Somewhere in the assumptions block sits an escalation rate for the utility price, applied every year of the term. The pattern to check is not the escalator's existence. California retail electricity prices have risen materially over the past decade, and a model that froze them at today's level would mislead in the opposite direction. The pattern is asymmetry: the avoided utility rate compounds upward while the machine's own cost lines, fuel, the service agreement, consumables, insurance, stand still or crawl.

Compounding is the quiet salesman. At five percent annual escalation a number doubles in roughly fourteen years; the arithmetic is illustrative, but it explains why so much of a twenty-year model's savings sits in years eleven through twenty, exactly the years no one can verify and the counterparty may not be present to stand behind. When the escalated line climbs and the offsetting cost lines do not, the out-year spread is a modeling choice wearing the costume of a forecast.

The asymmetry appears in every technology family, in mirrored forms. A fueled machine's model tends to hold gas flat while electricity climbs; the honest version escalates both from sourced, dated bases and shows the spread sensitivity plainly, because the project's economics are a bet on that spread whether or not the model admits it. An unfueled system's model has no fuel line to flatter, so the flattery migrates to the avoided rate itself: an aggressive escalator on the utility side, with inverter replacement and storage augmentation escalated gently or not at all. Same lean, different rows.

One more honesty note cuts against every long model, including the owner's own. Rate design changes. Over twenty years the structure of tariffs, not just their level, is likely to move, in directions no spreadsheet can source today. That is an argument for wide sensitivity ranges and shorter payback thresholds, not for anyone's point forecast. The discipline: every escalated line names its base, its source, and its date; both sides of the ledger escalate from evidence; and the sensitivity table shows the term economics at meaningfully lower and higher escalation. If the recommendation flips inside a plausible range, the owner deserves to see the flip in writing before signing.

Section 04The missing lines: standby, departing load, and the bill that arrives anyway

In the flattering version, the utility bill simply shrinks by the value of everything the machine produces. The bill a California site actually receives after commissioning is a more complicated document, and two families of charges are the most commonly omitted.

The first is standby. A site that installs generation and stays connected is asking the utility to stand ready to serve the full load when the machine is down, and utilities price that readiness. Pacific Gas and Electric Company's standard standby agreement, a tariff-book form, reserves standby capacity for exactly this purpose and bills the reserved capacity monthly at the rates of the applicable standby schedule.¹ The reservation charge is not a penalty. It is the price of the insurance that the model's own resilience narrative depends on, and a pro forma that credits the machine with outage protection while omitting the standing cost of the grid behind it is counting the benefit and hiding the premium.

The second family is departing load. California's restructuring era left a set of non-bypassable charges that follow load when it departs utility supply for onsite generation. PG&E's published guidance for customer-generation departing load lists six categories: public purpose programs, nuclear decommissioning, the Department of Water Resources bond charge, the power charge indifference adjustment, the energy cost recovery amount, and competition transition charges.² The utility's tariff book carries a dedicated schedule implementing the cost responsibility surcharge for departing customer generation.³ Two facts matter for model reading. Charges in this family can appear on a bill even where the customer is taking little or no utility energy. And exemptions genuinely exist, varying by technology and by the date the load departs. The honest model neither assumes the charges away nor assumes them in; it shows the determination for this site, this technology, this year, with the tariff sheets attached.

A third, smaller omission compounds the first two: the service that remains. After commissioning, the site's residual utility load has a different shape than the original account, and it may be billed under different provisions than the ones the model copied from last year's invoices. The one instrument that catches all three omissions at once is a post-project bill simulation: the utility invoice the site should expect in month thirteen, line by line, under the tariff that will actually apply, prepared or independently checked by someone whose fee does not depend on the equipment being purchased.

Section 05Incentives booked before they are earned

The incentive block is where models most often borrow certainty they have not purchased. Under current federal law as of this writing, the investment tax credit for qualifying clean-energy property is 30 percent.⁴ Statutory bonus adders exist for specific content, location, and program conditions, and every one of them carries its own qualification tests, documentation burden, and timing rules. An adder is not a default. A model that stacks adders into a headline credit without a written qualification path for each is not being optimistic; it is transferring qualification risk to the owner, because the purchase price does not refund itself when a credit fails to qualify.

Three disciplines keep the block honest. First, the base case runs at 30 percent for qualifying property and zero adders; adders enter only with a qualification memo from qualified tax counsel, and the model states who bears the loss if qualification fails. Second, tax attributes are worth face value only to an owner who can absorb them. Credits and depreciation monetize against tax actually owed, on the owner's actual schedule, under the owner's actual structure, and that is a conversation for the owner's tax advisors, not a cell the seller fills in. Third, program incentives sunset. California's Self-Generation Incentive Program is the local example: its budgets were authorized through 2025, and its legacy general-market budgets are closed to new applications as of this writing, so a 2026 model still crediting the program is citing history unless the program administrator confirms an applicable exception.⁵ Every incentive row deserves an as-of date, and a row without one should be treated as expired until shown otherwise.

Section 06Blended-rate savings math

The most common savings calculation in the field is also the least defensible: divide last year's utility spend by last year's kilowatt-hours to get a blended rate, then multiply that rate by every kilowatt-hour the machine produces. It feels rigorous because it starts from the site's own bills. It fails because no California business buys electricity at its average rate.

Federal data makes the point at the coarsest level: in May 2026, California's average commercial and average industrial retail prices sat roughly four cents per kilowatt-hour apart.⁶ Averages differ that much between whole sectors; within a single tariff they differ far more across hours, seasons, and billing components. A real commercial bill is energy charges that vary by time period, plus demand charges tied to the site's peak, plus fixed and rider charges tied to nothing the machine does. Onsite generation displaces energy in the specific hours it runs. It reduces demand charges only to the extent it is actually running, at output, at the moment the billing peak would otherwise have occurred. It displaces the fixed components not at all, and Section 04's charges can run the other way. Blended-rate math silently credits the machine with all three components at once.

The error is the method, not always the direction. For a genuinely firm machine running continuously behind a demand-charge-heavy tariff, tariff-correct arithmetic can show more value than the blend, because the blend dilutes demand savings the machine truly captures. For a resource that produces off-peak or intermittently, the blend flatters. Owners do not need to guess which way it cuts. They need the calculation done correctly once: twelve months of interval data, the applicable tariff applied bill by bill, savings decomposed into energy, demand, and fixed components, and every demand-charge credit tied to an explicit assumption about the machine's availability at the site's peaks.

Section 07Terminal value and the end of the story

The last years of a long model do quiet work, and four patterns recur. Residual value appears at end of term for equipment with no documented secondary market. Mid-life capital events, engine major overhauls, turbine hot-section work, fuel-cell stack replacement, battery augmentation, are smoothed into a service line or pushed outside the analysis entirely. Removal, disposal, and site-restoration costs are absent, even though lease and easement terms often make them the owner's obligation. And renewal-era service pricing is assumed at today's rates, though no counterparty has agreed to those rates for those years.

Honesty cuts both ways here too. Some equipment classes carry well-documented overhaul economics and real secondary markets, and an owner who reflexively zeroes every terminal line will undervalue a machine whose second decade is financeable. The standard is documentation, not pessimism: a residual value is either supported by a market someone can name, or it is zero; removal is a priced cost unless a counterparty has contractually taken it. The cleanest single test of the whole model is to recompute the return with terminal value at zero and removal carried as a cost. If the investment case changes materially, the project is being sold on its epilogue, and epilogues are the least certain pages of any twenty-year story.

Section 08The owner's line-by-line checklist

The six patterns compress into one table and ten asks. None requires proprietary data or an engineering staff. Each requires only that a specific document exist before anyone signs.

1. Unmerge the energy line.Annual output is capacity times hours times availability times degradation. Require the four factors separately, reconciled to the maintenance calendar in writing.
2. Escalate both sides.Whatever compound rate lifts the avoided rate, the fuel, service, and consumable lines carry their own sourced escalators, and the model is re-run symmetrically.
3. Demand the month-thirteen bill.A line-item simulation of the post-project utility invoice under the tariff that will actually apply, not a shrunken copy of last year's.
4. Price the grid you keep.Identify the standby provisions that apply and the monthly cost of reserved capacity before crediting any resilience value to the machine.
5. Confirm the departing-load determination.Which non-bypassable charges apply to this site, this technology, this year; which exemptions are claimed; tariff sheets attached.
6. Rebuild incentives at the statutory base.30 percent for qualifying property and zero adders until each adder has a written qualification path reviewed by qualified tax counsel.
7. Replace the blend with the tariff.Savings decomposed into energy, demand, and fixed components, computed from twelve months of interval data against the actual rate schedule.
8. Tie demand savings to coincidence.The model states its assumption about the machine running at the moment of billing peaks, and that assumption survives the maintenance calendar.
9. Put capital events inside the term.Overhauls, stack replacements, augmentation, and consumables at quoted prices, in the years they occur, not amortized out of sight.
10. Re-run the ending.Terminal value documented or zero, removal priced as a cost, and the return recomputed on that basis. If the answer moves materially, the epilogue was carrying the plot.

Section 09What the missing lines tell you

Read correctly, a vendor model is a genuinely useful document, just not as an analysis. It is a map of the seller's own expectations. The lines present at favorable values are where the seller expects scrutiny. The lines absent are where the seller expects none, which makes the omissions the most informative rows in the spreadsheet. An owner who runs the ten asks above is not being difficult; they are converting an offer into terms both sides can defend, and sellers with strong machines generally survive the conversion well. Disciplined reading is a compliment to good equipment.

None of this calls for cynicism about any technology. Engines, turbines, fuel cells, linear generators, solar, and storage each hold territory where they are the honest answer, and each is routinely oversold outside it. The model is where overselling becomes visible, one cell at a time, to any owner who insists on documents. That insistence is the method of this entire series: sourced numbers, dated facts, both sides of every ledger, and a decision the owner can still defend after the seller has gone.