Building the model is one thing; making it as precise and accurate as is reasonable is another — and then
there’s using it to find insights and drive decisions. This layer is about both: how you pressure-test
the levelized cost — ~$800/t for the example plant, give or take a ±30%
band — until you know how far to trust it, and how you turn a single number into a basis for deciding
what to do next.
This layer runs in five chunks:
4.1 — Validation — before any sensitivity, confirm the number is even
plausible: rebuild it by a different route, set it beside real comparables, and compare the cost
structure percent-by-percent.
4.2 — Sensitivity analysis — turn the single number into a
model you can flex. Sweep one input, then two, then coherent scenarios, and find the two or three
drivers the whole answer rides on.
4.3 — Using the analysis — act on what you found: feed the
drivers back to the science as your next experiment, and chase accuracy over precision.
4.4 — Framing the figure — decide what each figure has to
show and whether the cases in it are comparable at all, before any of it reaches a dashboard or a
deck.
4.5 — Picking the form — match the chart to the question and
to what the points actually are: independent observations, or repeated runs of one model.