Rule 17 of 38 · Chapter III — Make Failure Visible

Every error needs an owner

Why this rule exists

An error path with no owner is an error path nobody handles. When it is unclear whether the caller, the callee, or some middleware is responsible for a failure, each assumes another will deal with it, and the failure falls through the gap. Ownership is what turns a raised error into a handled one.

In practice

For every error a function can produce, decide who resolves it: the layer that can retry, the layer that can inform the user, or the top-level handler that logs and gives up cleanly. Make that boundary explicit in signatures and types. When you raise an error, know which layer is meant to catch it, and don't catch one you can't actually act on.

When it doesn't apply

Truly unrecoverable conditions — out of memory, a broken invariant — have one owner by default: the process boundary that logs and exits. You don't need a bespoke handler for every catastrophic case.