Mags

The challenge

Running scheduled and long-lived jobs reliably means handling the unglamorous cases: retries, partial failures, visibility into what ran and what did not, and an operator interface that makes a stuck job obvious rather than silent.

Scope

A full product rather than a feature — scheduling, execution, persistence, an operator-facing interface and the deployment around it.

What the factory did

• The requirement was written up as a specification with acceptance criteria before implementation began.
• Work was decomposed into a dependency-aware ticket graph so independent pieces could be built in parallel.
• Each ticket ran in an isolated environment on its own branch, with a working preview attached.
• Verification ran against the acceptance criteria, not only against the tests the implementer chose to write.

Where the engineers were involved

• Architecture decisions reviewed before the ticket graph was executed.
• Every merge approved by an engineer, with the approval recorded against the ticket.
• Security-sensitive paths reviewed by hand rather than accepted on test results alone.

Outcome

Running in production and used daily. The requirements, decisions, tickets and reviews that produced it remain queryable alongside the code.