Guide · Layer 4
Written for: seed / Series A scientist-founder
Layers 4.1 to 4.3 got you a number you can defend and a short list of what it rides on. Before any of that moves into a dashboard or a deck, decide what each figure needs to show. For most readers the figure is the only part of the analysis they will actually look at.
Misleading TEA figures usually aren’t the wrong chart type. More often they compare cases that were never comparable in the first place, and formatting won’t fix that. So pick the question before the chart.
A figure that works makes three things obvious: what is being compared, where each number came from, and how much confidence to place in it. Four things to settle before you pick a form.
Decide what the reader should learn, then choose the chart and the title around it. Write the claim as a sentence first — this figure shows that ___ — because comparability is relative to that sentence. The same data can support one claim and be unusable for another.
“Which route has the lowest cost in the base case?” wants a dot plot across the named routes. “At what power price does our route stop beating the incumbent?” wants a response curve or a break-even contour. Same model underneath; different figures, because the questions are different.
Start by writing down what one point represents: one plant, one project, one published source-case, one model run. The test is whether the same sentence can describe every point in the figure. If it can’t, split them into separate groups.
Before setting cases beside one another, check whether they share the same technology, geography, scale, cost year, system boundary, and units. Electrolyzer costs from different countries, technologies, and build years usually can’t be compared as reported. Work through the differences in this order:
If you’re testing whether capex falls with scale, keep every project at its actual capacity and plot capacity against cost. Capacity scaling is useful for transferring a reference cost into your model; it isn’t a way to remove the scale difference before you study economies of scale. Same trap with cost year: converting to a common price year is fine, but keep the original estimate year visible. Inflation adjustment puts money on a common basis; it doesn’t remove differences that came from technological learning.
Observed plants, published estimates, model assumptions, calculated outputs, sensitivity cases, and future projections are different kinds of evidence. They can share a figure, but they shouldn’t be presented as one undifferentiated series. A modeled 2050 scenario shouldn’t look like another operating plant. Labels, shapes, colors, or separate groups all work; what matters is that the reader can tell at a glance what each point is. Keep the source in the figure, the caption, or an accompanying table.
Different evidence types sharing a figure is often the point — a validation chart deliberately puts your model beside real comparables, and that comparison is the whole reason the figure exists. What they shouldn’t share is a series that makes them look like the same kind of evidence.
Say what every range means. A ±30% engineering estimate band is not a statistical confidence interval, and drawing them the same way invites the reader to treat them as one. Round the result to match the evidence, carry the accuracy class with the headline number, and label where the range came from.
A single value can still hide real variability. An electrolytic route shown at one grid-emissions factor can look clearly lower-carbon than a capture-based route, even though the comparison flips with the electricity source. When a driver can change the conclusion, show the range or plot the result against that driver — a sharp single number here is false precision doing real damage.
With the question fixed, the comparison fair, the evidence types distinct, and the ranges labelled, the form is nearly decided for you. 4.5 — Picking the form →