Code Quality
Trace the live behavior far enough to find the earliest divergence and the owner of its meaning. A symptom at the UI, cache, or adapter does not establish that layer as the right place for policy.
Useful questions include whether a legitimate mutation can bypass a check, and whether a rule would survive replacement of the layer that currently implements it. Map both producers and consumers when changing an API, schema, or boundary.
Prefer a direct durable change. Prepare structure only when the current shape blocks the change or its verification. Include required contract fallout; nearby cleanup earns its place through the problem being solved.
For stateful work, consider applicable identity, interleaving, commit, retry, and copied-state failures. The stateful scan and causal change notes offer deeper prompts. A regression should observe the changed contract with an expectation independent of the implementation; use a focused probe when that is more informative.
An independent mapper or specialist can expose a mistaken causal model. Pass source facts and label your interpretation as a hypothesis. See delegation.