This article explores the creation of a neural network that replicates the behavior of a simple if statement in programming. The goal is to produce the same output as a conditional statement that returns 10 when a variable ‘a’ is greater than 4 and 12 otherwise. The approach involves constructing a fixed list of arithmetic operations rather than training the network. The process begins by defining a ‘flag’ variable that acts as a decision maker, using two ReLU (rectified linear unit) functions to determine whether ‘a’ is greater than 4. The flag is then used to compute the final result through a weighted sum. The article explains how this method ensures the network behaves consistently for all integer inputs between 0 and 100. It also highlights the use of a Rust-based implementation called handwired-rs, which translates ordinary code into neural network operations. The code includes a test that verifies the network’s correctness by comparing its output to the original if statement for all 101 possible inputs. The result confirms that the network performs as expected, with no errors detected. The article concludes by discussing the implications of using exact arithmetic in floating-point operations and how the design ensures precision without rounding errors.