If / Else Merge

If / Else Merge (cond_merge) is the single interior node that makes an If/Else conditional compound work. It has three inputs — condition, then, else — and one output: $\text{result}=\text{is\_true}(\text{condition})?\text{then}:\text{else}$.

One node, two contexts. Inside a Conditional compound (execution_mode = Conditional) the engine evaluates the condition first, prunes the untaken branch's exclusive sub-region from the graph, and only then runs the rest — so when this node executes, only the taken input is present and it simply forwards it. The untaken branch's nodes never run at all: an expensive computation or a would-divide-by-zero path is skipped, not computed-then-discarded (Simulink If-Action-style lazy pruning). Anywhere else (plain dataflow) both branches were already computed eagerly upstream and this node still forwards the taken one — so a bare drop on the canvas degrades gracefully into a scalar select. (For per-sample routing, use the switch node instead.)

The condition is reduced to exactly one boolean for the whole branch, by a single shared rule so the engine's prune decision and this node's runtime select can never disagree: a Boolean uses its value; an Integer or Scalar is truthy when $\ne 0$ (a NaN counts as false); a Signal is reduced to true iff any sample is non-zero, with a warning that per-sample conditions belong on Switch; an absent or non-numeric condition is false (take the else branch, with a warning for the non-numeric case).

The taken value is forwarded verbatim — its type, unit and uncertainty ($\sigma$) all carry through unchanged; the merge adds no physics of its own. An empty or absent taken branch emits a $0$ scalar and a warning. Every port is optional, precisely because the engine intentionally leaves one of then/else absent after pruning.