An insufficiently appreciated paradox in the origin of life is that the replication of information-carrying molecules requires the molecules to be very specifically shaped; but such specific molecules are hard to produce without natural selection. We demonstrate and investigate this problem by building a physical model of self-replication out of specifically shaped plastic pieces with embedded magnets, which float around on an airhockey type table. We use a mechanism known as template replication, which works by the joining of complimentary strands, roughly analogous to the biological replication of DNA, except without the involvement of enzymes. Building a physical rather than a computational model forces us to confront several issues that have analogues in the microscopic, chemical world. In particular, in order to achieve a low mutation rate we must reduce as much as possible the formation of incorrect sequences, which can happen both spontaneously and as a result of strands joining in a misaligned way. The latter results in ever-lengthening sequences in a process known as the “elongation catastrophe”. We present an overview of our design process, illustrating the many interdependent adaptations that had to be made to the pucks’ shapes in order to solve these problems while maintaining a high rate of template replication. The chicken and egg question is how, in the pre-biotic world, could template replication be achieved without the presence of enzymes that require template replication in the first place? By building a real physical model a new answer to this question is suggested. We propose that early pre-biotic monomers required structural specializations that reduced the rate of formation of incorrect sequences, without the need of an encoded enzyme.