Test each component of a compound schema independently — compound failures give no diagnostic

Why This Is a Rule

A compound schema — "I believe A AND B AND the interaction between A and B produces C" — fails as a single unit. When C doesn't appear, you don't know whether A was wrong, B was wrong, or the interaction model was wrong. The compound failure provides zero diagnostic resolution.

Testing each component independently first — verify A alone, verify B alone, then verify the interaction — produces diagnostic clarity at every step. If A fails independently, you know the failure point without needing to test the compound. If A and B both pass independently but the compound fails, the interaction model is the failure point — a diagnosis impossible from compound testing alone.

This is the same logic as unit testing before integration testing in software: unit tests isolate failures to specific components, making them debuggable. Integration tests on untested units produce failures that could live anywhere in the stack.

When This Fires

• Before trusting any multi-part plan, strategy, or belief system
• When a complex schema produces an unexpected failure and you need to diagnose why
• During validation of any model that combines multiple independent claims
• When "something is wrong but I can't tell what" describes the diagnostic situation

Common Failure Mode

Testing only the compound: "I'll try the whole strategy and see if it works." If it fails, you've learned only that "it didn't work" — not which component broke. You either abandon the entire strategy (losing the components that were correct) or re-run it hoping for a different result (no diagnostic improvement).

The Protocol

For compound schemas: (1) Decompose into atomic components — each independent claim or assumption. (2) Test each component independently: does this claim hold on its own? (3) Components that pass → validated atoms. Components that fail → repair or replace before compound testing. (4) Only after all atoms pass: test the compound (the interaction between validated components). Compound failures now diagnose to the interaction model, not to unidentified component failures.