An unfalsifiable schema is a belief system, not a model — reclassify it and flag for replacement
When a schema cannot specify any observation that would falsify it, classify it as a belief system rather than a testable model and flag it for replacement or constraint.
Why This Is a Rule
A testable model makes predictions that can fail. "Experienced engineers produce fewer production bugs" is testable — you can measure bug rates by experience level. A belief system explains everything retroactively but predicts nothing testably. "Everything happens for a reason" is unfalsifiable — any outcome can be accommodated by finding a "reason" after the fact.
The distinction matters operationally because testable models improve through feedback (failed predictions trigger revision), while belief systems are immune to feedback (every outcome confirms them). Operating on an unfalsifiable schema means you're running on autopilot without a correction mechanism — the schema can't tell you when it's wrong because nothing counts as wrong.
Reclassifying an unfalsifiable schema as a "belief system" isn't necessarily discarding it — some belief systems serve valuable psychological functions (meaning-making, motivation). But they should be clearly labeled as belief systems rather than masquerading as testable models. The label change prevents you from treating unfalsifiable beliefs as if they were evidence-based conclusions.
When This Fires
- During schema audits (Five-step schema audit: list rules, source origins, find successes/failures, rate confidence, check evidence, Document consequential schemas with falsification conditions — unfalsifiable schemas are dogma) when testing falsifiability
- When a schema has never been wrong (suspicious — all schemas encounter exceptions)
- When you can't articulate what evidence would change your mind about a belief
- When a schema seems to explain everything without predicting anything specific
Common Failure Mode
Confusing flexibility with unfalsifiability. "People respond to incentives" is falsifiable (you can test specific incentive-response predictions) even though it's broad. "The universe is guiding me" is unfalsifiable (any outcome is interpreted as guidance). The test: can you specify, in advance, an observation that would make you abandon the schema? If no → unfalsifiable.
The Protocol
For each schema in your inventory: (1) Ask: "What observation would falsify this?" (2) If you can specify a concrete, observable falsification condition → testable model. Keep and test. (3) If you cannot specify any falsification condition → belief system. Reclassify: label it explicitly as a belief system, not a tested model. (4) For belief systems: decide — does this serve a valuable function despite being untestable? If yes → keep but constrain (don't use it for decisions where testable models are available). If no → replace with a testable alternative.