Anticipate failure modes

Realistic failures include data leakage, duplicate rows across splits, a non-representative validation set, class imbalance hidden by accuracy, selecting a model after hundreds of test-set checks, and distribution shift after deployment.

Classify each failure as prevent, detect, contain, or recover. Prevention is strongest when a hard invariant is possible: schema validation, access control, data-split isolation, or admission limits. Detection needs an observable signal and owner. Containment limits blast radius with tenant boundaries, read-only tools, canaries, budgets, or circuit breakers. Recovery needs a tested fallback, rollback, re-index, or human queue.

Avoid a vague instruction such as “be careful.” Write a tripwire: a metric threshold, validation error, unexpected version, or forbidden action. Then state the response. If the response is “retry,” explain why the failure is transient and why retrying cannot duplicate a side effect or amplify overload.

A spam classifier has 99% training accuracy and 84% validation accuracy. Reducing tree depth moves those to 93% and 91%. The second model is preferable because its estimated generalization gap shrinks from 15 to 2 percentage points.

The worked number is generalization gap = training score − validation score = 0.99 − 0.84 = 0.15. State the unit and denominator whenever you report it. A percentage without a denominator can conceal a tiny sample; a latency without a percentile can conceal slow users; a similarity score without a labeled task can conceal irrelevant neighbors. Compare the observed value with a threshold chosen before seeing the final test result.

Now test the tempting shortcut. Suppose the team optimizes only the most visible metric. The result may look better while the system becomes less trustworthy. The reason is concrete: Too little capacity underfits real structure; too much fits noise. Strong regularization lowers variance but can add bias. More training data usually helps, but only if it represents deployment conditions. Repeatedly tuning on the test set quietly turns that test set into training data. This is why the decision record must include both the intended gain and the tolerated regression. If the tolerated regression is unknown, the change is not ready for a consequential workflow.

• Prefer a measured baseline over a persuasive demo.
• Keep versions, inputs, and thresholds reproducible.
• Separate syntactic success from semantic correctness and authorization.
• Escalate or abstain when evidence falls outside the contract.
• Re-evaluate when data, traffic, models, providers, or user goals change.

These rules turn the topic into an engineering decision rather than a slogan. They also make disagreement productive: another person can challenge the assumptions, rerun the evaluation, and reach a documented conclusion.

Choose the failure with the largest combination of likelihood and impact. Inject it in a test environment without weakening production controls. Capture the first observable symptom, the alert that should fire, the component that contains the damage, and the recovery action. Then remove one safeguard and predict how the blast radius changes before running again. The lesson is not that every failure can be detected from model text. Strong designs enforce invariants outside the model and preserve enough evidence to distinguish bad input, component failure, policy refusal, and ordinary low-confidence output.