HCD.4:4.3.1 - Test and Revise the Model without Rewriting the History

For a relation that can change the next decision, design an observation whose plausible outcomes favor different accounts. Specify the task, means, relevant first action and inference before interpreting its result. An answer that both accounts predict supplies little discrimination. HCD.11 returns the observation and its limits; it does not make the proposed categories true merely by scoring them.

Distinguish three updates. New evidence may locate this person differently within an unchanged model; it may require a different description, condition or relation in the model; or it may only require changing a task or available help. If a supposedly higher performance appears while its asserted prerequisite fails under comparable conditions, examine the prerequisite claim, the observations and the model’s grain. Do not relabel the inconvenient attempt as noise simply to preserve the ladder. If the apparent contradiction comes from an inaccessible task or changed help, repair that comparison before changing the proficiency account.

Use learner explanations and work products to detect a strategy missing from the proposed categories. Where the categories collapse different mechanisms, split them; where two categories never change a relevant inference, consider merging them. Reconsider task validity, scoring and alternative explanations along with the model. Small local evidence can justify a reversible pathway change while leaving a population-wide progression unvalidated.

Keep the earlier observation, its conditions and the model interpretation used at the time. Record the revised interpretation with its reason. Revising a model is a change in an account; claiming that the person learned, forgot or transferred requires evidence about the person. HCD.14 uses this distinction when the model revision also changes the remaining learning arrangement.