Your State Machine Should Be Data, Not a Workflow Engine
Most systems that model a process reach for one of two extremes: a big switch on an enum, or a full workflow engine. Both hurt, in opposite directions.
The problem
A hard-coded state machine means every new status is a release. A workflow engine means a second runtime, a second mental model, and a designer nobody on the team wants to own.
The split that worked
Put what in the database and how in code:
- Which statuses exist, and which moves are permitted between them — rows.
- What must be true before a move is allowed — named predicates in C#.
A transition row references its preconditions by key:
public interface ITransitionGuard
{
string Key { get; } // matches a key on the transition row
GuardResult Evaluate(GuardContext context);
}Adding a rule becomes a row plus, at most, one new class. Adding a status becomes a row and nothing else.
Why it matters
The engine never reads your status constants. Those exist for guards, seeders and tests. That's what keeps the table authoritative — the moment code branches on if (status == Approved), the database has stopped being the source of truth and you have two of them.
One caution: this only works if the table is validated. A transition naming a precondition that no longer exists must be caught loudly, not skipped silently. More on that later in the series.