Guide · Layer 4
Written for: seed / Series A scientist-founder
The model gives you a number; the point now is to turn it into something you can flex — put in different assumptions, see different outcomes, and watch the relationships between the inputs and the answer. Sweep at three widths.
Run a one-way sensitivity on each input — flex it across its credible range, hold everything else fixed, record the output swing — then rank the swings in a tornado chart. The longest bars are your drivers: the few inputs the whole answer rides on. Three things to watch as you sweep:
The analysis is the range, not the wiggle: leverage = how steeply the output responds × how far the input can plausibly move. A blanket ±10% on every input is the signature beginner mistake — it buries the real drivers under well-known inputs that can’t actually move; draw each range from its own uncertainty (a price’s market history, a proxy’s error band). Expect 2–3 inputs to carry the answer — here, the power price, the capital, the energy intensity. And read the chart as leverage, not likelihood — the bars aren’t additive, and the spread isn’t a probability distribution. (The capacity factor Layer 3 flagged draws a short bar here — its real leverage lands in the multivariate scenarios below.)
Some drivers only bite in combination. Flex two at once and read the result as a small grid or surface — cost per tonne against, say, power price on one axis and capacity factor on the other. It shows where the pair crosses a threshold — a break-even contour — that neither one-way sweep reveals on its own. Reach for it when two inputs plausibly move together, or when a decision hangs on where a line sits rather than on the length of a single bar.
One-at-a-time analysis misses combinations — the real downside is usually several inputs going wrong together. A scenario is a coherent bundle of inputs moved at once, describing one internally consistent world. Two kinds matter here:
Coherence is the discipline: move the drivers together in a way that could actually happen — a stressed grid plausibly hands you a low capacity factor and a high power price at once — and don’t stack every input’s independent worst value, which manufactures a world that can’t exist and a scare number nobody believes. Keep the set small and probability-free: three cases communicate, thirty don’t, and the spread is not a confidence interval. Compare routes on one shared basis. And hold two ranges without blurring them — an honest internal “low-low” case you run the company against, and a legitimate upside case you put to investors. Just stay clear which is which — Layer 5’s honest-headline rules hold.
Treat the three widths as a loop, not a one-pass exercise: a flat sweep sends you back to check the wiring, a long bar sends you back to tighten a range or bundle a scenario, and each pass sharpens which inputs actually deserve the attention.