The Fuel Line:
Modeling Gas Price
Exposure Honestly

In 2022 the U.S. benchmark natural gas price averaged $6.45 per million Btu, the highest annual figure since 2008; two years later the same benchmark posted the lowest inflation-adjusted annual average ever recorded. A multi-year power model that carries fuel at one flat number has made a large, silent bet on its most volatile input, and most owners are never told the bet was placed.

Section 01Every fuel-consuming path has a fuel line

When a large energy user prices its credible paths to power, most of the serious on-site candidates consume fuel. Reciprocating engines, gas turbines, microturbines, linear generators, and fuel cells differ in combustion, efficiency, emissions profile, and maintenance burden, but they stand in the same line at the same counter: each buys gas, year after year, at whatever the market charges. Over a long operating life, cumulative fuel purchases for a continuously running machine rival and often exceed the capital cost of the machine itself. Capital is quoted, negotiated, and fixed at signing. Fuel is repriced by a continental commodity market every day for the rest of the project's life, and no buyer, seller, or consultant controls it.

The standard modeling failure is the single-point assumption. A lifetime cost table is built on gas at one number, escalated at one guessed rate, and the result is presented to a board with decimal precision. The decimals are real; the foundation is not. Where the point came from is usually unstated, and whether the conclusion survives a different point is usually untested. That is not a forecast. It is an argument wearing the costume of arithmetic.

The practice persists because it is useful, and it is useful to everyone except the owner. A low flat gas price flatters every fuel-consuming machine, and the parties who sell them. A high flat gas price flatters every path that buys no fuel: the utility-service case, solar, storage, and the parties who sell those. The direction of the error tracks the direction of the compensation. Technology neutrality on fuel price does not mean choosing a moderate number. It means refusing to let one number decide.

This paper sets out the discipline we apply in study work: build the delivered price from its components, anchor the central case to dated forward indications, band it against realized history, report the price at which the recommendation flips, and apply identical treatment to the electricity side of the meter. A companion question, whether gas service physically exists at the site with capacity to serve a new machine, is a separate screen with its own paper in this series. Price is this paper's subject.

Section 02What the recent record actually shows

These are not stress-manual scenarios. They are consecutive entries in the published record of the most liquid natural gas benchmark in North America, from the federal government's own price series. In 2022 the Henry Hub spot price averaged $6.45 per million Btu, the highest annual average in both real and nominal terms since 2008 and more than 53 percent above the 2021 average, the fourth-largest one-year increase in the series; the year it was measured against, 2021, had itself posted one of the three larger increases.¹ Within 2022 alone, daily prices ran from $3.46 in November to $9.85 in late August.¹

Then the whipsaw. By 2024 the same benchmark averaged $2.21 per million Btu, the lowest annual average in inflation-adjusted dollars ever reported, and the fall from 2022 was the largest two-year decline on record.² Even inside that record-low year, the tails printed in both directions: a January cold snap produced a daily price of $13.49, roughly six times the year's average, and the four lowest daily prices ever recorded occurred in the same calendar year.²

Any modeling convention should be tested against that record. The most common convention, a sensitivity of plus or minus ten percent around a chosen point, fails immediately: recent annual averages span roughly a threefold range, and single days within a record-low year printed at multiples of that year's average. A band narrower than realized history is not caution. It is decoration.

Volatility also cuts both ways, and an honest paper says so. The 2022 spike punished buyers who were consuming gas unhedged. The 2024 trough punished anyone who had locked multi-year supply at 2022 levels. The record does not teach that gas is expensive, and it does not teach that gas is cheap. It teaches that the distribution is wide and that both tails have recently happened, which is precisely the information a single-point model deletes.

Section 03The price at your meter is not the benchmark

On December 21, 2022, the Henry Hub benchmark stood at $6.14 per million Btu. On the same day, the three major trading hubs serving the western United States, including the Northern California city-gate, cleared above $50.00 per million Btu, an average of $48.12 over the benchmark, and the Northern California and Oregon hubs recorded their highest prices since December 2000.³ The federal analysis of that event named the drivers: pipeline maintenance in West Texas cutting flows westward, Pacific-region storage roughly 30 percent below its five-year average, and cold weather that lifted regional residential and commercial gas consumption by 23 percent in the first three weeks of the month.³

That day carries the entire lesson about basis, the difference between the national benchmark and the regional hub that actually prices a site's molecules. Basis has its own drivers: pipeline capacity and outages, regional storage position, regional weather. In most months it is a modest adder. Occasionally it is the whole story. A California fuel model built on the national benchmark alone is a model of a meter that does not exist.

The price a site actually pays is a build-up: benchmark commodity, plus regional basis, plus utility transportation and surcharges under the applicable tariff, shaped by how the site buys. Utility-supplied customers inherit wholesale movements by design; the state regulator's own explainer of Northern California gas rates notes that the procurement component of the retail rate resets monthly with the wholesale market.⁴ Larger sites commonly hold transport-only utility service and buy the commodity from third parties, in which case the supply contract, indexed, fixed, or blended, decides which market movements reach the invoice. Each layer belongs in the model as its own line, with its own source and its own date.

Section 04Forward indications are inputs, not promises

For the commodity layer there is one continuously observable, dated, arm's-length reference: the exchange-traded futures strip on the U.S. benchmark, quoted in monthly contracts extending years forward and priced every trading day by parties with money at stake. As a central anchor it has a property no consultant's opinion can match: it is nobody's opinion. In study work we treat the strip, read on a stated date, as the default central case for the benchmark layer of the build-up.

What the strip is not is a promise. It is the set of prices at which buyers and sellers will transact today for delivery later, and realized prices depart from prior strips routinely, in both directions. Distant months trade thinly. The strip prices the national benchmark, not the California city-gate, so basis and transportation still have to be added from their own sources. And a strip read in March is a different input from the same curve read in August; the as-of date is part of the input, not a formality.

Official outlooks earn the same treatment. As accessed on the date of this paper, the federal Short-Term Energy Outlook carries the benchmark at $3.67 per million Btu for 2026 and $3.49 for 2027, and names weather as the main uncertainty around that path.⁵ The agency revises the outlook monthly. That posture, dated, sourced, and expected to be revised, is the correct posture toward every forward number in an energy model, including the ones in this paragraph.

Three rules follow in practice. First, every forward input in a study carries its as-of date on the page where it is used; an undated forward number is treated as missing. Second, forward indications set the central case only; the band comes from realized history and scenario logic, never from the curve's own apparent confidence. Third, when the inputs age past the decision they were gathered for, the model is rerun before capital moves. A model that has outlived its inputs is not conservative. It is stale.

A single-point fuel assumption is not a forecast. It is a decision already made, dressed as an input.

Section 05Bands, flip points, and the discipline of wide inputs

The honest representation of fuel price is a set of cases, not a number. In study work the minimum is three: a central case built from the dated strip, dated basis, and the current tariff; and low and high cases anchored to realized history, no narrower than the span of recent annual averages, widened further when the machine's life is long. The source of every band is stated in the document, and each site derives its own numbers from its district, its tariff, and its actual supply options.

The most useful single number the fuel analysis produces is the flip point: the delivered gas price at which the recommendation changes. If the paths under study reorder only at prices far outside any plausible band, fuel risk is real but not decisive, and the study says so. If they reorder inside the recently realized range, the recommendation is fragile, and the study says that too, on the first page and in plain language. An owner who knows the flip point can watch the market and understand what it means for the decision. An owner holding a single-point model can only hope.

Escalators deserve particular suspicion. A percent-per-year fuel escalation, guessed at signing and compounded across a long life, quietly dominates any lifetime figure it touches. The honest treatment shows year-one economics separately from lifetime economics, labels the escalator as an assumption, and runs the flip-point arithmetic on the escalator as well as on the level.

Correlation is where sophistication and honesty have to meet. In a wholesale power market where gas-fired plants frequently set the marginal price, electricity and gas prices tend to move together. For an owner weighing purchased power against self-generated power from gas, the two exposures partially offset, which softens the worst-case spread in both directions. But the offset is weaker at the retail meter than at the wholesale node, because California retail rates also carry wildfire, transmission, and distribution costs that do not move with the commodity. A model that treats the retail tariff as a pure gas play overstates the hedge; a model that treats the two prices as independent overstates the risk. A study should state which assumption it uses and why, because both extremes are wrong.

Hedging, finally, transforms risk rather than removing it. Fixed-price physical supply and financial instruments exist, for finite terms, at market cost, with credit and collateral conditions attached. The honest model shows hedged years as hedged and every later year as banded, and it never extends a short fixed term across a long machine life. Contracted renewable fuel supply, where a site pursues it, generally trades one structure for another, longer fixed terms at different price levels, and the study states the term and the level rather than blending them into one flattering average.

Section 06The same honesty, applied to electricity

Every abuse described above has a mirror image on the other side of the meter. A model that bands gas carefully while holding the retail electricity tariff flat for twenty years has quietly favored the utility-service path. A model that escalates the tariff with confidence while gas sits at a single point has quietly favored on-site generation. Both look rigorous. Each stresses only the side of the meter its author is not selling.

On the electricity side, the level is a fact and the trajectory is not. The federal Electric Power Monthly puts California's average commercial retail price at 24.10 cents per kilowatt-hour in May 2026, against a national commercial average of 13.54 cents.⁶ Where the tariff goes from there is decided in regulatory proceedings: published and knowable one rate case ahead, genuinely uncertain across a project's life, in both directions. Capital programs argue for continued upward pressure; affordability interventions and new load spreading fixed costs argue the pace can change. The discipline is identical to the gas side: cite the tariff as of a date, band the trajectory, and report the electricity-side flip point with the same prominence as the gas-side one.

The table is the neutrality. Every row carries its own honest liability: the fuel-consuming rows are exposed to everything in Sections 02 and 03; the non-fuel rows escape the tariff only in part; the utility row buys from a wholesale market that the same commodity moves. No row wins on paper until the bands are applied to a specific site's load shape, air district, gas service, and deadline, which is why the comparison is study work and not a brochure table.

Section 07What a defensible fuel line contains

Collected in one place, the anatomy:

• A delivered-price build-up. Benchmark, basis, transportation, and contract structure, each with its own source and date, none assumed away.
• A dated forward anchor. The central case tied to a forward strip read on a stated date, labeled as an indication rather than a prediction.
• Bands no narrower than recent realized history. Or a stated reason for the exception, in writing.
• The flip point. The delivered price at which the recommendation changes, stated in plain language where the reader cannot miss it.
• Escalators labeled as assumptions. With year-one economics shown separately from lifetime economics.
• Hedge terms at their actual tenor. Never silently extended across the machine's life.
• Symmetric treatment of electricity. Its own dated sources, its own bands, its own flip point.

None of this is technically difficult. It is commercially inconvenient, and that is the honest explanation for its rarity: wide bands lose sales that narrow points win. A party compensated by which answer wins cannot afford the band. A party compensated for the analysis cannot afford anything else. Our fee does not change with the answer, which is the only reason this paper can be as blunt as it is.

The fuel line is where an energy model most often stops being analysis and becomes advocacy. The remedy costs a few pages: components, dates, bands, and flip points, on both sides of the meter. An owner who insists on those pages, from us or from anyone, will still face a wide distribution of possible futures. The difference is that the decision will have been made facing that distribution, rather than facing a single number chosen by someone with something to sell.