Rule of Law

Rule of Law is a logical predicate language and a tool that allows you to verify assumptions about data and relationships in your database that are otherwise hard to check or enforce.

Think of Rule of Law as a last line of defense, to provide a safety-net for cases that normal database constaints cannot cover.

Example

Here's an example:

rule "All completed orders must have a completion date"
forall orders o:
  o.status = "COMPLETE" => o.date_completed != NULL

Read the above as "for every order in the DB it should hold that if its status is COMPLETE, then it must also have a completion date". Note that this expression says nothing about non-COMPLETE orders.

One could also state the opposite: "for every order in the DB it should hold that if it has a completion date, then it must also be in COMPLETE status":

rule "All orders with a completion date must be complete"
forall orders o:
  o.date_completed != NULL => o.status = "COMPLETE"

Since this arrow holds both ways, these two rules can be combined into a single rule with an equivalence relation:

rule "All completed orders have a completion date, and all others do not"
forall orders o:
  o.date_completed != NULL <=> o.status = "COMPLETE"

Rationale

Codifying these rules as logical statements serves multiple purposes:

1. Share common system knowledge. By making the rules explicit, they become an expression of intent.
2. Explicit documentation.
3. One place to look. Since all rules can be stored in the same place, it's easy to find.
4. Proactive monitoring. Since these rules are self-validating, we can periodically run them as checks against real production data and alert as soon as a counter example is found.
5. Test cases. By running the rules as part of the CI / testing phase, you can catch broken assumptions as they happen.
6. Guide code reviews. Because the rules are explicitly stored and versioned alongside code, they allow us to link to a rule when reviewing code.
7. Part of the developer workflow. While working on new code, formulate a one-off rule and verify it quickly against real data.