[Domain name]

Copy this file to docs/development/domains/<name>.md and answer every section. The suite fails if a heading is missing, if the controlled cost share carries no percentage, or if that percentage is below 15. This file is scaffolding and is skipped by those checks.

One paragraph: what the laboratory searches for, and what exists at the end of the search that does not exist today.

Value function

Test 0, and it is free. Can the customer buy an article today that performs the same function, in any form, at any price? If so the value is loss reduction, which is bounded below by zero loss and therefore saturates — and the incumbent, who owns that cost line, can move process architecture far more cheaply than materials. Seven candidates died here.

Write the value function down. V = C − K/x, or a reciprocal sum 1/V = Σ 1/v_i, is a rejection. State the form as a threshold — the article exists above the line and does not exist below it — or as multiplicative into a quantity nobody currently sells.

Physics-limit sufficiency

Test 2. R ≥ 3. Take the discoverable property to its physical ceiling. Divide the improvement it delivers by the improvement required for parity with the best architecture available to the incumbent. Current practice is the wrong benchmark, because the incumbent will upgrade.

The threshold is 3 so the margin can absorb the pioneer-plant penalty: 51% of pioneer process plants never reach 85% of design capacity, and class-2 estimates come in 1.28× over.

Architecture ratio

Test 3. A < 0.3, and this check has caught seven of seven. Divide the value of the best zero-discovery flowsheet change by the value of the materials axis at its ceiling. Ask what the incumbent's process engineers could do to the same cost line with no discovery at all, and price it.

Search the flowsheet literature and the licensor patent record, and name what you found.

The screen. Answer it first, and answer it with arithmetic. What percentage of the paying customer's cost does the discoverable property govern? Derive it from public cost structure. State the denominator explicitly — cost of the delivered product, project capital, cost per tonne — because the three falsified candidates all sounded large against the wrong denominator.

Show the working, including the numbers you had to assume. Below 15% the domain is rejected.

Process maturity

Does a mature incumbent process already exist that this would improve? If it does, the domain is rejected, because in a mature process the material axis is the cheapest axis and incumbents harvested it decades ago.

State which condition holds: the product cannot currently be made at all; the incumbent route is indirect enough that a direct route is a different process; a recently cheap capability opened an unsearched space; or a funded buyer has no supply route. Give the evidence.

Annual tonnage

World annual tonnage of the addressable output, with a source. A market measured in tens of tonnes per year cannot carry a venture at any price per kilogram.

Attribution distance

How many transformation and integration layers sit between the materials advance and the entity that pays? Name each layer. Above three, the value is captured by someone else.

Computational regime

SOLVED, PARTIAL, or BLIND. Only PARTIAL qualifies — theory predicts direction and cannot rank candidates reliably. State what simulation or machine learning contributes, and how much. The published base rate is an acceleration factor with a median around six, degraded two to three times by measurement noise, and flat beyond roughly ten to twenty experiments per dimension.

Measurement identity

Is the number the literature reports the number that sets value? Name both. This fails constantly: one field reported gravimetric equilibrium selectivity on powder in binary feed while plant cost was set by volumetric working capacity and mass transfer on a formed bead in real multicomponent feed.

State whether the cited record was measured under conditions the commercial article will see.

Fast screen predicts slow truth

What is the cheap measurement, what is the expensive one, and what evidence says the first predicts the second? Without that, the facility is a machine for generating confident error at speed, and the domain is rejected however good the science is.

State the sign of the correlation in the variable that must change at scale. A correlation that merely exists is insufficient: one falsified candidate had selectivity falling with the pressure its plant required, so a better bench number predicted a worse plant.

Note whether the expensive measurement can be truncated as well as scheduled around, and say what the bridge costs to build — decision D13 makes it the facility's first deliverable.

Capital intensity

Discovery value per unit of output against capital charge per unit of output. Ownership is justified when the first exceeds the second. State both numbers and the assumptions behind the capital charge.

Note where the margin sits and whether the business stays specialty, since integration that moves a company from specialty to basic chemicals costs roughly a third of enterprise value per dollar of earnings before anything is operated.

What the facility does

The experiment, its cost, its wall-clock time, the fast and slow split, the shared instruments, and the sample custody chain. Name real instruments. A domain where one instrument does everything is a worse fit, because it would not exercise facility-scale capability.

What would have to be true

Numeric gates, each with a threshold, a cost, and a duration. Write the hard stop here, before the work starts.

The honest case against

The strongest argument that this is wrong, stated as its proponent would state it.