Surprise = automatic schema review — do not wait for the next scheduled cycle
Treat surprising outcomes as automatic triggers for schema review rather than waiting for scheduled validation cycles, as surprise signals that at least one schema in your stack has drifted from reality.
Why This Is a Rule
Surprise is the real-time signal that your model and reality have diverged. You expected X, reality delivered Y, and the gap — the surprise — is information about where your schema is wrong. Waiting for the next scheduled review to investigate the gap allows the drift to persist and potentially produce more failures.
Scheduled reviews catch gradual drift. Triggered reviews catch sudden changes. Both are necessary, but triggered reviews have higher diagnostic value because the surprise provides a specific, time-stamped data point: "At [time], in [context], I expected [X] and got [Y]." This specificity makes the investigation targeted rather than exploratory.
The rule makes surprise an automatic trigger — not "consider reviewing when surprised" but "review now, triggered by surprise." The automaticity prevents the common response of explaining away the surprise ("that was unusual, it won't happen again") and deferring review to a scheduled cycle where the diagnostic specificity has faded.
When This Fires
- When any outcome surprises you — positive or negative
- When a prediction you made turns out wrong
- When a familiar process produces an unfamiliar result
- Any moment where reality diverged from your expectation in a way that made you pause
Common Failure Mode
Normalizing the surprise: "That was weird, but probably a one-off." Surprises that are explained away without investigation are missed opportunities for schema correction. The explanation ("one-off") might be correct, but the investigation is how you find out. Without the investigation, you're choosing comfort over calibration.
The Protocol
When you experience surprise: (1) Notice it immediately: "I expected [X] and got [Y]." (2) Do not explain it away. (3) Investigate: which schema produced the expectation? What assumption in that schema does the surprise contradict? (4) If the schema needs revision → revise and generate new predictions (Schema revisions must generate new predictions, not just explain away the failure). If the surprise was genuinely anomalous → document it and monitor for recurrence. The investigation takes 5-10 minutes; deferring to the next scheduled review means the specific diagnostic signal will have decayed.