Rating engines, rules versus models, referral workflows, and rate-change governance.
3 questions in Insurance.
In a filed environment the rate is a regulated artefact rather than a configuration value, so the change is versioned by state and effective date, applies to business written from that date rather than to everything, and must leave every existing quote and policy reproducible against the version that priced it.
Model the rating algorithm as an ordered set of steps over versioned rate tables, pin every quote to the exact rate version and input snapshot that produced it, and keep eligibility and referral rules outside the price calculation so an underwriter decision never changes the number.
It has to be available at quote time for a known share of risks, obtainable as it stood historically so a backtest means anything, defensible as a factor rather than a proxy, and stable enough that the vendor cannot change your filed rate for you. Everything else is a modelling detail.