E.9.DA:11 - SoTA-Echoing and source use
| Practice question | Exact source and status | Selected payload and disposition | Changed E.9.DA locus | Qualification and smallest reopen condition |
|---|---|---|---|---|
| How much structure should the first decision result expose? | Nogueira, Silva, and Conte, One Size Fits All? An Empirical Comparison of ADR Templates regarding Comprehension, Usability, and Ease of Adoption, 2026 preprint — current empirical comparator for ADR-form usability. | Adapt. The study’s controlled comparison favours Nygard’s concise and objective form overall, while participant feedback identifies MADR’s advantage for structural detail and specific requirements. E.9.DA therefore keeps a short ordinary result and adds detail only for a named reliance. Reject one universal record shape or the inference that the longest template is the most adequate decision. | §§4, 4.6, and 5: ordinary result first; reliance-bearing extension second. | Qualified on 2026-08-15 for the reported student experiment and five templates. Reopen these loci if a stronger practitioner study reverses the form trade-off or shows that a named E.9.DA use needs a different minimum. |
| How should a reviewer treat fluent, reconstructed, or generated rationale? | Zhou, Li, Liang, et al., Using LLMs in Generating Design Rationale for Software Architecture Decisions, 2025 preprint — current empirical stress line for rationale completeness and misleading additions. | Adapt, with a domain boundary. In the reported 100-problem study, generated rationale had incomplete recall and a small but real share of potentially misleading arguments. E.9.DA does not generalize those rates to all DRRs; it uses the result to reject fluency, source volume, or recovered prose as completeness evidence and to require one bounded search for an important omitted question. | §§4, 4.4, 4.4a, 5, and CC-E9DA-0b/9. | Qualified on 2026-08-15 for generated software-architecture rationale, not all human-authored decisions. Reopen if better direct evidence changes the omission/misleading-risk answer or supplies a more effective low-cost completeness test. |
| Which architecture-description distinctions may an architecture-facing DRR reuse? | ISO/IEC/IEEE 42010:2022, Software, systems and enterprise — Architecture description — current-standard reference, not SoTA by official status. | Adopt narrowly. Reuse the standard’s explicit separation between an entity’s architecture and an architecture description, and its boundary that the standard does not define the architecting process or recording medium. Reject treating the standard, an architecture description, a viewpoint, or a model as the DRR adequacy method or as the architecture decision itself. | ArchitectureSourceAndViewLossClosure, the architecture-impact slice, and Relations. | Qualified to the published 2022 edition checked on 2026-08-15. Reopen only if a relied-on distinction changes; a later edition number alone does not make the standard a better adequacy method. |
| How should an actionable finding connect present condition, desired condition, and next move? | Sadler, Formative assessment and the design of instructional systems, 1989, and Hattie and Timperley, The Power of Feedback, 2007 — retained feedback lineage, with current FPF use carried through E.22 and E.23 rather than claimed as new SoTA. | Adapt as lineage. Keep the gap, evidence locus, first repair or drafting action, and reopen condition together so a finding changes the next move. Reject praise, a score, or a condition label without an action-changing diagnosis. | §§4.4a, 4.5–4.7, the worked case, and finding-related checks. | Retained because the desired/current/next-action distinction still answers this narrow feedback question and is used by E.22/E.23. Reopen only if a better feedback result changes that action structure, not because the sources are old. |
| How should several adequacy dimensions and trade-offs remain visible? | Qin et al., A survey on Quality-Diversity optimization: Approaches, applications, and challenges, Swarm and Evolutionary Computation 100 (2026) 102240 — current multi-dimensional optimization overview; MCDA and Pareto practice remain design lineage. | Adapt only the preserved-dimension lesson. E.9.DA keeps distinct coordinates, adjacent-value reasons, and visible trade-offs; it does not average them or import a QD archive, algorithm, fitness function, or selection mechanism into DRR review. The effective-floor rule is noncompensatory. | §§4.3–4.5, calibration, Consequences, and CC-E9DA-8/10/11. | Qualified on 2026-08-15 for the current overview’s quality-plus-diversity separation. Reopen if the selected dimensions no longer lead to distinct repair questions or a better decision-evaluation practice preserves the same distinctions at lower use cost. |
| What prevents decision-adequacy values from replacing decision usefulness? | Karwowski et al., Goodhart’s Law in Reinforcement Learning, ICLR 2024 — current theoretical and empirical proxy-optimization branch; Goodhart and Campbell remain lineage. | Adapt. Treat the ordinal value as an imperfect proxy claim, ask whether an intended value or protected trade-off became worse, and stop value-directed repair when it improves the visible score without strengthening the selected answer or first action. Reject all-5, 5-defensible, source-count, or checklist-completion targeting. | Problem failure 9; §§4.4a and 4.7; CC-E9DA-8/11; the Goodharted-DRR anti-pattern. | Qualified on 2026-08-15 as a proxy-risk result, not as an assertion that DRR review is reinforcement learning. Reopen if better proxy-risk work changes the early-stop or protected-value answer used here. |
The current-best source set spans empirical decision-document use, rationale-completeness risk, multi-dimensional evaluation, and proxy optimization. ISO/IEC/IEEE 42010 is deliberately kept as a narrow current-standard reference; the feedback, MCDA, Pareto, Goodhart, and Campbell traditions remain lineage where the current rows or neighbouring FPF patterns carry the present answer. A new publication date, official status, citation count, or popular template does not by itself reopen any row.