Every number closes, or the engine says why it can't.

A stream comes in with a declared composition. The engine runs it through the physical chemistry that applies — neutralization, phase change, combustion, transport losses — and checks whether mass, energy, and charge balance at the end. If they do, every reported figure traces back to either a balance that closed or a declared input carrying its own source.

If a step needs a constant nobody has sourced — a settling velocity, a heat-transfer coefficient, an activity model outside its valid range — the engine does not substitute a textbook-plausible guess. It reports that line as blocked, names the missing source, and lets everything else in the dossier stand on its own. A partial answer you can trust beats a complete one you can't.

A stream can create value four different ways. The engine evaluates each route independently, and each is built in its own right — not a shared formula skinned four different ways.

Feasibility studies get killed in review because a number in them can't be traced back to anything. Every figure this engine reports is one of two things: the output of a balance that actually closed, or a declared input with its own citation attached — never a number typed in because it looked about right. That is the property a technical reviewer is actually checking for, and it's the one a lookup-table tool can't offer.