Concept

Scenario (assumption bundle)

analytical

Overview

A scenario is a coherent bundle of input assumptions changed together to describe one internally consistent state of the world — a route choice (gray / blue / green), or an optimistic / base / pessimistic case — with the output recomputed under the whole bundle. Where a one-way sensitivity moves a single input at a time, a scenario moves a self-consistent set at once, so it captures the combinations one-at-a-time analysis cannot.

Body

What it is. A scenario is a named set of simultaneous input choices that hang together, each answering “what if the world is this way?” The output is computed once per bundle, giving one result per scenario. The defining feature is that the inputs in a bundle are chosen to be mutually consistent — they describe a world that could actually occur, not an arbitrary mix of values.

Two common kinds.

Route / structural scenarios — different process configurations or technology choices, like the gray, blue, and green ammonia routes. These differ in model structure (which unit operations exist, where the boundary falls), not merely in parameter values.
Condition scenarios — the same process under different external conditions, bundled into low / central / high (optimistic / base / pessimistic) cases. Each bundles a consistent set of prices, performance, and financing assumptions.

Why bundle inputs. A scenario captures correlated and joint moves that a one-way sweep structurally cannot: the realistic best or worst case is a combination of inputs, not the end of any single tornado bar. It also lets discrete alternatives that are not a continuous range — route A vs. route B — be compared head to head.

Consistency is the whole discipline. A bundle is only worth computing if its inputs are jointly plausible. Pairing a favorable value on one input with a contradictory value on an input correlated with it — cheap power and a high capacity factor when the two are coupled through the same grid conditions — describes a world that cannot exist, and its output is meaningless however reasonable each value looked alone.

Built from the drivers. A useful scenario set is usually assembled by moving the few high-leverage drivers — found by sensitivity analysis — to a consistent low or high state together, while leaving non-drivers at base. Flexing parameters the output is insensitive to adds dimensions without adding insight.

Limits & typical error

Incoherent bundles — the signature error. Assembling a “worst case” from every input’s worst value independently, without checking which inputs are correlated, builds a state that cannot occur and overstates the spread. The honest worst case respects the correlations among inputs, and is usually far less extreme than the independent-worst stack.
A few discrete cases sample the space sparsely. Three bundles are three points in a high-dimensional space; the real outcome can fall outside all of them, and the gap between low and high is not a range the three points actually bound. Labeling them “best/worst” implies limits they do not establish.
Scenarios carry no probabilities. They are deterministic what-ifs with no likelihood attached, so the central case is not an expected value and the outer cases are not confidence bounds. Reading a base case as a mean, or a pessimistic case as a P90, imports statistics the method does not provide.
The base case absorbs optimism. Hopeful assumptions tend to settle into the base case, and a scenario set built around it inherits that center — so a “downside” that is really an honest base, and a base that is really the optimistic case, mislabels the entire set.
Structural scenarios must share a basis to be comparable. Route scenarios change the model itself, so comparing gray vs. green only means something when both are on the same system boundary, cost basis, and output definition; differing bases turn a route comparison into an apples-to-oranges error.
The bundle count explodes combinatorially. Varying k drivers over m states each gives mᵏ possible bundles; past a small number the set stops communicating a clear comparison, and the interpretability that is a scenario’s main advantage over probabilistic sampling is lost.

Mini-example

The three ammonia routes are the running example’s canonical route scenarios: gray, blue, and green share the identical synthesis loop and differ only in the hydrogen source and the CO₂-capture block (the locked structure from the reference sheet). Each is a coherent bundle — green = {electrolytic H₂, ~10 MWh/t power at ~$40/MWh, near-zero carbon intensity on renewables}; gray = {SMR hydrogen, ~30–40 GJ/t natural gas, gas-price-driven cost, CI ~1.6–2.4 t CO₂e/t} — and the three are comparable because all sit on the same boundary and the same $/t levelized-cost basis.

Within green, a condition scenario bundles the drivers coherently: a downside case sets a low capacity factor (~~0.45) and a high power price (~~$55/MWh) together, because a stressed-grid world plausibly produces both at once. That bundle lands well above ~$800/t — worse than either input’s one-way swing alone, which is exactly the combined effect a scenario exists to show.

Separately, to show an incoherent bundle: a “worst case” built by stacking green’s worst value on every input independently — lowest capacity factor and highest power price and highest capex and worst intensity — describes a world that does not occur, since the high-renewables conditions that depress a plant’s capacity factor often lower power prices rather than raise them. Ignoring that correlation manufactures a downside no real state produces.