C.35:11 - SoTA-Echoing
| Source or practice line | Adopt, adapt, or reject | Concrete C.35 locus changed | Boundary and currentness |
|---|---|---|---|
| MBSE query and view generation | Adapt query results as discovery while separating the claim-bearing result, represented object, representation, and publication-side availability. | Adds query or extraction basis, separate observed, inferred, and unknown content, covered and unexplored model region, uncertainty, validation, and C.30.AD / C.30.ASV exits. | Query or view output is not architecture, realized structure, or proof. Reopen when result identity, model edition, query rule, viewpoint, represented object, coverage, or relied-on availability changes. |
| Graph grammars and model transformations | Adapt rule-governed transformation as the transformation branch. | Adds exact source and result objects, transformation trace, preserved structure, lost structure, and C.34 preservation exit. | Grammar or transformation output does not prove adequacy, feasibility, or realization. Reopen when transformation rules, source object, result object, trace, or constraints change. |
| DSM, MDM, and modularization practice including Jiang and Luo, arXiv:2604.28018 | Adapt modularization and LLM-assisted DSM work as discovery. | Adds extraction basis, observed dependencies, inferred clusters, unknown functional-bearer semantics, coverage, uncertainty, validation, and C.31 plus C.32 exits. | Cluster, partition, or MDM slice is not candidate architecture adequacy. Reopen when relation matrix, covered region, modularity objective, functional prior, validation, or solution pool changes. |
| Multi-objective NAS and Sukthanker et al., arXiv:2402.18213 | Adapt multi-objective search as a generative-proposal source. | Adds constraints, proposed neural organization, known omissions, validation needs, search criteria, bearer boundary, eval return, and C.32 admission condition. | A Pareto point or neural graph is not holonic architecture adequacy. Preservation is claimed only against an exact declared baseline. Reopen when search space, constraints, criteria, hardware target, proposal content, or eval trace changes. |
| DSE, QD, OEE, NQD, and evolutionary architecture practice inherited through C.32 | Adapt retained alternatives and stepping-stone pressure as candidate-input practice. | Strengthens candidate-generation input, result-use return, archive exit, front exit, pool-policy exit, and C.32 coordination. | These practices do not make C.35 a second candidate-set admission rule. C.18, C.19, and G.5 define archive, front, and pool policy; C.32 defines candidate-palette admission. |
| AI-assisted architecture design and AI-assisted ADD | Adapt generated decompositions, relation graphs, and decision proposals through the generative-proposal branch. | Adds constraints, proposed organization or claim content, known omissions, source-label recovery, validation needs, and candidate-admission boundary. | An LLM proposal, ADD suggestion, benchmark trace, or agent consensus is not decision authority, evidence sufficiency, realization, or architecture adequacy by itself. |
Sapunov, Theory of Code Space, and code-agent architecture-map practice | Adapt partial-observability mapping through the discovery branch. | Adds observation or extraction basis, separate observed, inferred, and unknown content, confidence, covered and unexplored regions, active-passive comparison, and validation. | A code-agent map, JSON probe, benchmark score, dependency F1, invariant F1, or active-passive gap is not architecture adequacy, internal-state proof, safe-change authority, evidence sufficiency, gate passage, or release authority. |
| GonzoML neural-network architecture intake | Adapt neural architecture operation language as source-label recovery for discovery or generative proposals. | Adds recovery for dataflow change, routing, gating, memory placement, cache placement, block substitution, pruning, distillation, NAS, ablation, and compute, memory, and latency trade-offs without choosing a branch by label. | Neural-network labels, ablation gains, pruning masks, distillation success, and search outputs remain source cues until the branch-specific basis, bearer, affected characteristic, and next architecture claim plus its required rule are recovered. |
C.35 rejects the popular shortcut that a generated result, Pareto point, cluster, graph, or diagram is architecture because it looks useful. Recover the exact result first; add representation or publication details only when they matter; then state the intended use, missing condition, forbidden overread, and return.