NSTD.Preface:5 - Evaluation, typical failures and consequences

Before relying on a narrative, ask what the reader needs to recover, whether the account supports that recovery, and which remaining uncertainty changes the next action. Inspect an inferred cause or attribution against its source. For a draft or generated candidate, use the defect to choose a repair; acceptance is not a prerequisite for evaluation.

NSTD.6 applies C.2.8 to the account, its expression and a specified reader or observer. State the selected semantic relations, correctness criterion, preparation, help and budget as needed for the question. A qualitative comparison can suffice. A count also needs its grain and selected set. Retain which relations were recovered: equal counts may support different actions.

An expert walkthrough, an actual response and a formal estimate establish different things. One supported response does not establish unaided transfer or reliability across a population. Current product quality, fidelity to a named source and evidence of how the product was made are separate questions. Missing manufacturing notes do not prevent diagnosis of a visible defect.

Typical repairs follow those distinctions: qualify a cause inferred from sequence; restore a hidden viewpoint or condition; attach a prerequisite where the learner needs it; compare a generated scene with continuity constraints; or retain an adequate explanation when further work would add little. If repeated improvement is worthwhile, use E.22 to frame an unsettled question and E.23 to compare the change with protected characteristics, cost and risk in view.

The gain is an account whose content, order and use are easier to examine and change. The cost is the comparison and any necessary reader or subject work. Additional forms, probes and evidence are justified by what their result can change. A consequential permission, evidence, assurance or affected-party claim needs its own grounds; narrative quality alone does not supply them.