When a binary label hides multiple distinct reasons, split it into separate evaluable dimensions
When a binary classification hides multiple distinct failure modes or reasons within a single bucket, decompose it into separate dimensions that can be evaluated independently.
Why This Is a Rule
Binary classifications compress multi-dimensional reality into yes/no buckets, and the compression discards the information needed to improve. "Failed" is binary, but the reasons for failure are multi-dimensional: the candidate failed because of technical skills, or communication, or culture fit, or unclear job requirements. The binary "failed" label hides which dimension caused the failure, making it impossible to improve the process that produced the failure.
When multiple distinct reasons map to the same binary outcome, the binary label becomes actively misleading: it treats fundamentally different problems as the same thing. "Rejected" applications where the candidate was underqualified require different interventions than "rejected" applications where the candidate was overqualified or "rejected" applications where the role was mis-scoped. Same label, completely different causes, completely different fixes.
Decomposing into separate dimensions — evaluating each reason independently — restores the information that the binary compression destroyed. Each dimension can now be tracked, analyzed, and improved independently.
When This Fires
- When a binary label (pass/fail, yes/no, approved/rejected) is used for decisions with multiple possible causes
- When analyzing failure data and finding the binary label provides no actionable insight
- During any classification design where items in the same bucket have different root causes
- When "failed" items are lumped together despite having nothing in common except the outcome
Common Failure Mode
Adding more binary labels instead of adding dimensions: "Failed-Technical" and "Failed-Culture" are still binaries — they're slightly more specific but still compress information. The fix is dimensions: score technical skills 1-5, score communication 1-5, score culture fit 1-5. Dimensions preserve the information; binary labels destroy it.
The Protocol
When a binary classification hides multiple reasons: (1) List all distinct reasons that could produce the same binary outcome. (2) Convert each reason into an independent dimension with its own scale or criteria. (3) Evaluate items on each dimension separately rather than collapsing to a single binary. (4) The multi-dimensional evaluation costs slightly more per item but produces actionable data about which dimensions to improve — which the binary label never could.