PROJECT / SAMPLE

The safest rewrite was not rewriting it.

Instrumentation and behavioral tests made a fragile legacy service understandable enough to repair incrementally.

Date
2026-04
Type
software / reliability
Status
fictional fixture
Published
prototype only

Problem

A small essential service had accumulated years of patches, undocumented dependencies, and manual restart rituals. Nobody understood it well enough to change it confidently.

Constraints

  • Existing behavior was the only reliable specification.
  • The service could not tolerate a long replacement project.
  • Several external dependencies were no longer maintained.

Observed

  • Most incidents originated in two recoverable dependency failures.
  • Obsolete paths carried disproportionate operational risk.
  • A rewrite would have recreated years of undocumented edge cases.

Work

  • Instrumented the service before rewriting any path.
  • Reconstructed its runtime and dependency map.
  • Added tests around observed production behavior.
  • Removed obsolete paths and isolated failure modes.
  • Migrated the highest-risk components incrementally.

Decisions

  • Treat production behavior as evidence, not an inconvenience.
  • Remove risk in descending order rather than replacing the system at once.
  • Document recovery before changing architecture.

Result / SAMPLE

The service became boring enough to operate and safe enough for ordinary maintenance without a risky replacement program.

  • 37 incidents/quarter → 2
  • MTTR: ~95 min → 11 min
  • full rewrite avoided

What we did not do

We did not rewrite a system we did not yet understand. Instrumentation bought more certainty than new code would have.

Technical index

observability / behavioral testing / legacy systems / incremental migration

← all project records · bring us a real problem →