systemic-issue-triage

Classify every issue by root class before touching code: superseded by an in-flight change, duplicate of a known class, real new bug assigned to a root-cause cluster, feature request, or unclear. Issues sharing a root cause get one fix that closes all of them with named test evidence.

Root-cause clustering groups two or more issues sharing the same underlying defect into a single fix at the root, rather than N separate patches. Each closure is proven by a named test, and the goal is a net-negative line delta that shrinks the system.

Apply the over-engineering test: if a fix adds a state, verb, config flag, gate, or parallel representation of existing truth, redesign it. The correct fix usually deletes or relaxes something, such as reducing 13 states to 5 or turning a dead-end into a one-line exit message.

The test may have been taught to agree with the defect. Check whether the commit introducing the bug also modified a test helper, fixture, or corpus to satisfy it, or whether a test row asserts the defective behavior as intended. Remove such tests rather than leaving a bug pinned in place.

One-by-one fixing treats each symptom independently, causing the system to accumulate states, flags, and machinery exponentially. It also misses duplicate root causes, so the same defect gets patched repeatedly while the system grows harder to maintain.