Guide · Layer 4
Written for: seed / Series A scientist-founder
The analysis earns its keep when you act on it — and what it’s good for is relative comparison and prioritization between decisions, not handing you a precise absolute number. Be clear-eyed about what you can and can’t learn at this stage: it’s a tool for choosing between two paths, not for pinning a figure to four significant places. Two ways the drivers pay you back.
The drivers tell you where the economics are actually decided — the next experiment to run, the assumption to pin down, the parameter worth improving. Sort them on a second axis the tornado doesn’t show — controllability:
The inverse rule matters as much: don’t refine a non-driver, however shaky it feels — sharpening a parameter the answer is insensitive to improves your confidence, not your decision. Run this loop and the TEA stops being a report and becomes a research roadmap.
Your biggest lever should be one you control, not one the market hands you — if the longest bar is the grid price, the honest move is to say so out loud and go hunting for a controllable lever to put R&D behind. And when a team asks “is 95% good enough, or do we need 98%?”, let the model answer: push the parameter to each value and read the headline. Whether the juice is worth the squeeze is better settled with a sensitivity than by gut.
Push a recovery from 95% to 98%, or a price across its range, and the headline can move more than the input did — small changes routinely drive big swings in the answer. The lesson isn’t to chase the input to another decimal; it’s to aim at accuracy — a number that’s defensibly in the right place, with an honest range — over precision, a tidy figure that happens to be wrong. When you catch yourself polishing a digit the next sensitivity would wash out, that’s false precision, and it’s the thing to resist.
The deeper method gives every input a distribution and samples them all at once, for a full distribution of the output — a P10/P50/P90 on your cost per tonne — capturing the interactions that one-way sweeps and a few scenarios both miss. It earns its keep when a counterparty’s diligence wants a quantified risk view, when input interactions genuinely dominate the answer, or when you’ve moved past go/no-go into formal risk analysis. The cost: every input now needs a defended distribution, not just a range — and a Monte Carlo output looks rigorous whether or not the distributions behind it are real, which is exactly where its false-precision risk lives. A P50 to four significant figures built on a dozen guessed distributions tells you less than a deterministic ~$800/t ±30% with three named drivers. At the maturity anchor, deterministic sensitivities plus a few coherent scenarios are deliberately the right tool.
Once you know what moves the answer and how robust it is, the last layer makes it travel: Layer 5 — Communication Layer, where the headline number and its few drivers go into a deck and in front of investors — without giving away the trade secret that produced them.