Test schemas at the smallest possible scale first — observe results against pre-stated predictions

Why This Is a Rule

Schemas that feel correct in your head may fail in reality. The only way to know is to test them — and the most efficient testing is at the smallest possible scale, where the cost of failure is minimal and the learning is maximum.

"Smallest possible scale" means: one customer instead of a market launch. One conversation instead of a company-wide policy change. One week's experiment instead of a permanent commitment. The small-scale test preserves the learning value (does the schema predict reality?) while minimizing the cost of being wrong.

Pre-stated predictions are essential because without them, you'll interpret any outcome as consistent with the schema (confirmation bias). "I predicted that [specific outcome] would occur under [specific conditions]" — then observe whether it actually did. The gap between prediction and reality is the diagnostic signal.

When This Fires

• Before acting on any belief that hasn't been tested against reality
• Before scaling a small success into a larger commitment
• When a schema feels theoretically sound but has never been reality-tested
• During any decision where the stakes are high enough that testing at small scale first is worthwhile

Common Failure Mode

Testing at too large a scale: "I'll test my management philosophy by restructuring the entire team." If the schema is wrong, the restructuring damages the team before you learn. Test at minimum viable scale: apply the philosophy in one 1:1 meeting and observe results against your prediction.

The Protocol

Before consequential action based on a schema: (1) State your prediction: "If [schema] is correct, then [specific observable outcome] will occur when [specific action] is taken." (2) Design the smallest-scale test: what's the minimum action that would produce observable evidence for or against the prediction? (3) Execute the test. (4) Compare actual results to your pre-stated prediction. (5) If prediction holds → increase scale cautiously. If prediction fails → revise the schema before scaling. The small-scale test costs hours; the untested large-scale action costs months.