Behavior Characterization

Detect what the system does today without judging whether it is ideal.

Observable signals

• Return values.
• Exceptions and error codes.
• Side effects and state changes.
• Persistence writes/reads.
• Published events.
• Relevant logs.
• External calls.
• Edge conditions and branch behavior.
• Implicit business rules.

Every behavior claim needs code evidence or must be marked as a hypothesis.

Characterization method

• Write an assertion expected to fail, run it, and read the actual value from the failure message; anchor that value as the expected result. The test documents what the code does, not what it should do.
• Characterize only the zone the plan will touch (targeted characterization), not the whole unit by default.

Bugs found while characterizing

Characterize the bug as-is: the test anchors today's wrong output. Never fix it in the same step — record it, finish the safety net, and schedule the fix as a follow-up behavior change to run with the net in place.