Overview — Model and software design

A quantitative model that only exists in a notebook cell cannot be tested, bumped or shipped. This topic turns scientific-Python numerics into software: small instrument and curve classes with a clear split between market data and model parameters, numerical tests that use closed forms and invariants, a reproducibility record beyond "I set a seed", and a profiling habit that vectorises the hot line instead of guessing.

By the end of this note you should be able to:

• design a small quant object (discount curve, bond, or European payoff) with composition over deep inheritance;
• write tests that lock closed forms, finite-difference bumps, parity identities and seeded randomness;
• record enough environment and data identity to rerun a result;
• find a slow line with a simple harness and decide whether a rewrite is worth it.

1. Contents
2. The big picture