| Fluency benchmark value as quality | Smoothness replaces structure recovery. | Inspect the selected relations and evidence. Return a missing basis or repair a demonstrated loss through NSTD.1 through NSTD.3; fluency alone establishes neither a high nor a low amount. |
| Style repair as precision repair | A nicer wording pass is treated as sufficient while relation kind, quality sense, language-state threshold, or coarsening loss remains hidden. | Lower LanguageStatePrecisionAndCoarseningFit for the unresolved defect. Clarify it directly when its meaning is determined, or use the relevant precision, coarsening, explanation or language-state pattern for the remaining question. Re-evaluate the repaired rendering before claiming improvement. |
| Prompt loop as improvement | The worker keeps regenerating more engaging drafts without a named object version, allowed change slice, protected trade-offs, or re-evaluation. | Open E.22 when needed, route the repair to E.23, and re-evaluate the changed version through NSTD.6; otherwise retain it as a candidate without claiming improvement. |
| Evaluation theft | Quality result is used as evidence, assurance, or gate. | Keep NSTD.6 as evaluation; route wider use to A.10, B.3, or gate owner. |
| Evaluating a different object | A judgement about one narrative is used to rate its generator, source archive or whole course. | Select the evaluation for that object. A script or generated draft can still be assessed as a narrative when it expresses one. |
| Reconstructed history as evidence | A plausible route inferred from a good product is reported as the author’s actual construction or repair. | Retain the current-product result and mark the historical claim unestablished until its own evidence is obtained. |
| Characteristics as eval programs | Test scripts or automated checks are treated as characteristics. | Keep characteristics in A.19.ECS; automated evals are measurement or eval-program carriers under direct owners. |