Sensitivity Analysis: Which Inputs Move the Answer

Sensitivity analysis ranks inputs by how strongly they change a result, telling us where precision is worth buying and where it is not.

Not all inputs matter equally

A design has many inputs, but usually only a few control the headline result. Sensitivity analysis finds them. We vary each input across its plausible range and measure the response, producing a ranked picture of what actually drives Q_sci, fusion power, or peak field.

Local and global sensitivity

Local sensitivity is the derivative at the design point: useful and cheap, but blind to interactions and to behaviour far from the point. Global sensitivity varies inputs together across their whole ranges and attributes output variance among them. We use local analysis for quick guidance and global analysis before trusting a conclusion, because interactions can dominate near thresholds.

What sensitivity tells us to do

• Buy precision where it pays: measure or pin down the high-sensitivity inputs first.
• Stop polishing low-sensitivity inputs: extra digits there do not change the answer.
• Flag fragile results: if a small input change flips a conclusion, the conclusion is not robust and we say so.
• Design measurements: the FOAK instruments should target the quantities the design is most sensitive to.

For the burner (Aegis / MetroVolt), sensitivity analysis is part of why the plug regime is such a stark gate: performance is extremely sensitive to conditions that sit 166-830x beyond any operated device, so the result cannot be pinned down until those conditions can be reached and measured. Sensitivity analysis does not soften a gate; it explains why the gate is where it is. In the end, sensitivity analysis is what turns a long list of inputs into a short list of the ones that decide the answer, and it tells FOAK instrumentation exactly which quantities are worth the effort of measuring first.