Guide · Layer 2
Written for: seed / Series A scientist-founder
Layer 1 gave you the shape. This layer puts the physical numbers on it — every flow, split, and duty — so Layer 3 can put dollars on them. Get this layer right and the cost layer is mostly multiplication; get a flow wrong and no amount of careful costing saves it — you’ve put a clean price on the wrong number.
The work is one move repeated: take each box from Layer 1, boil it down to a handful of numbers that turn its inputs into its outputs and energy use, then make everything reconcile through the mass and energy balances. One habit runs through all of it: build the balance so its holes are obvious — the weak assumptions you can see are a lot less dangerous than the ones you can’t.
Keep it at the right altitude for the stage: deterministic, arithmetic-grade modeling — basic equations and engineering proxies, not a simulation. Two calls follow from that:
This layer runs in 5 activities:
Anchoring on the input is the setup we argue about most with teams. Bump a yield and you want output to move — pin the output instead and the same tweak silently resizes your whole upstream plant. A simple color convention keeps the line visible: blue = an input you chose, black = calculated, green = pulled from another sheet — anyone can see in a second what’s a lever and what’s a result.
Write each unit op as input → mechanism → output: a reactor is a conversion and a selectivity, a separator is split fractions, a compressor is a pressure ratio and an efficiency.
The parameter values are the analysis. The too-clean ones (100% conversion, zero losses) are exactly what a diligent reader checks first, so spend your care there. Get a rough number on every box before drilling into any one — a parameter you can’t actually pin down doesn’t sharpen the answer, it’s false precision with more cells to maintain. The question that walks a rough model toward an honest one: what’s the next-most-important parameter I haven’t pinned down yet?
Make the streams reconcile — mass in = mass out — around each box and around the whole plant:
Put your attention on the biggest streams — an error in a major feed moves equipment size and consumption both; a trace stream rarely changes the answer. When the algebra tangles, find a tie component — a species that rides straight through untouched backs out an unknown flow in one division. And lay the balance out inside each box (input → mechanism → output), not as one wide stream table; if you keep a stream table, make it display-only — every number calculated elsewhere and referenced in.
Most processes convert only part of the feed each pass and recycle the rest, so two conversion numbers live side by side:
Closing the loop needs a recycle (send unreacted feed back) and a small purge (bleed a little off so inerts don’t build up). Don’t solve the loop by iterating it to convergence — assume a steady run-rate and write off a fixed bleed. The model stays deterministic and easy to re-run.
One rule prevents the expensive mistakes here: size equipment inside the loop on the loop flow; buy feedstock on the overall conversion. At ~20% per pass the recycle runs several times the fresh feed — mix those two numbers up and your compressor is badly undersized, or your feedstock bill is overstated by the same factor. And the purge has a price: it carries good reactant out with the inerts, which is a real feed loss, not a rounding error.
Keep unlike energy carriers on separate lines — a MWh of electricity isn’t a GJ of fuel: different price, different carbon. And a duty in MW says nothing about whether the heat is usable — temperature level decides what can drive what. If one big exotherm can obviously raise steam for a heating step, net it off your energy intensity — then stop. Chasing the last few percent of heat integration is a later-stage job.
The step up replaces proxies with real phase equilibria, kinetic reactor models that predict conversion instead of assuming it, and formal pinch analysis for heat integration — usually inside the same process simulator Layer 1 flagged for solving loops. It earns its keep when a P/T-sensitive equilibrium genuinely won’t yield to a proxy, when conversion is the thing you’re trying to predict, or when detailed design or a counterparty’s diligence demands it. One middle path before the full climb: run the simulator once to calibrate a correction factor on the one box that needs it, bake it into the hand model, and drop the dependency.
Once the flows and duties are fixed, the next layer puts dollars on them: Layer 3 — Economic Layer, where each duty becomes an equipment cost and each consumption figure becomes a cost per tonne.