Appendix B — Publication Strategy
B.1 Overview
The expanded project scope supports three distinct publications, each building on the prior:
| Paper | Repository | Target Venue | Timing | Status |
|---|---|---|---|---|
| GVSETS 2026 | gvsets/ | NDIA GVSETS | Draft Mar 23, Final Jun 5 | Drafted, evaluation pending |
| Grammar Transposition | kebnf-to-tree-sitter | MODELS/SLE 2026 or SE Journal | Q3-Q4 2026 | Outline complete |
| SE Benchmark for AI | sysml-grammar-benchmark | INCOSE IS 2027 | Q3 2027 | Notional |
B.1.1 Publication Dependencies
GVSETS 2026 (Foundation)
│
├──→ Grammar Paper (Formal Rigor)
│ │
└─────────┴──→ Benchmark Paper (Validation)B.2 GVSETS 2026: AI-Augmented MBSE
Working Title: “Enabling AI-Augmented Model-Based Systems Engineering with the Model Context Protocol”
| Attribute | Value |
|---|---|
| Track | Digital Engineering / AI |
| Format | 8-page technical paper + presentation |
| Draft Due | March 23, 2026 |
| Notification | May 1, 2026 |
| Final Due | June 5, 2026 |
| Presentations Due | July 23, 2026 |
| Presentation | August 11, 2026 (Novi, MI) |
B.2.1 Key Thesis
MCP provides a standardized interface enabling AI assistants to interact with SysML v2 models stored in Git repositories, offering token efficiency via selective retrieval (estimated 80-97% reduction with L0/L1/L2 detail levels, pending benchmark validation), structured responses eliminating parsing ambiguity, and authoritative answers from tools connected to real repositories.
B.2.2 Current Status
The paper is drafted in gvsets/paper/main.tex with all sections present. The evaluation section (Section 5) contains placeholder data pending benchmark vignette execution (V1, V4, V5 from Section C.1). Eight TODO markers flag unvalidated quantitative claims that must be resolved before submission.
B.2.3 3-Condition Experiment Design
| Condition | Description |
|---|---|
| Baseline | All files concatenated into prompt (naive approach) |
| Vanilla MCP | Simple tool calls without optimization |
| Optimized MCP | Cache ID + Summary pattern, L0/L1/L2 tiered responses |
Critical path: Execute benchmark vignettes V1/V4/V5 against the Eve Mining Frigate model to replace placeholder data with measured results.
B.2.4 Relationship to Capstone
This paper establishes the foundation — demonstrating practical AI-MBSE integration and proof of value. The systems engineering artifacts in this capstone (SEP, SyRS, ADD, VVP, RTM) provide the methodological rigor backing the paper’s claims. The expanded ecosystem understanding from scope exploration (7 projects) enables articulation of meaningful future research directions.
B.3 Grammar Transposition Paper
Working Title: “Automated Grammar Transposition: Converting OMG KEBNF Specifications to Tree-sitter Parsers”
Target Venue: MODELS/SLE 2026 or Systems Engineering Journal
| Attribute | Value |
|---|---|
| Format | 10-12 page technical paper |
| Target | Q3-Q4 2026 submission |
| Repository | kebnf-to-tree-sitter |
B.3.1 Key Thesis
Formal specification grammars (KEBNF) can be systematically converted to practical parser generators (tree-sitter) with ~93% automation and documented semantic mappings. This enables reproducible grammar generation when specifications update, formal traceability from parser rules to specification sources, and a reusable methodology for any OMG KEBNF-based specification (OCL, Alf, future textual notations).
B.3.2 Unique Contribution
This is the first documented methodology for KEBNF → tree-sitter conversion. No existing literature addresses KEBNF specifically, despite OMG’s use of KEBNF across multiple standards and tree-sitter’s rapid adoption across major editors and platforms.
B.3.3 Paper Structure
- Introduction (~1 page): MBSE adoption driving need for SysML v2 tooling; gap between OMG specifications and practical parsers; contribution statement
- Background (~2 pages): OMG grammar specifications, KEBNF syntax elements (type annotations, property assignments, cross-references, semantic actions), tree-sitter architecture
- Methodology (~3 pages): KEBNF pattern taxonomy, conversion algorithm (parse → classify → transform → record mapping → emit), semantic mapping document design
- Implementation (~2 pages): Tool architecture (Chumsky parser → Mapper → tree-sitter emitter), technology choices, conflict detection approach
- Case Study: SysML v2 (~2 pages): Input corpus (640 rules across KerML + SysML KEBNF files), automation results by category, comparison with hand-written tree-sitter-sysml grammar
- Discussion (~1 page): Applicability beyond SysML (OCL, Alf), limitations, tree-sitter enhancement opportunities
- Conclusions (~0.5 page)
B.3.4 Current Status
The kebnf-to-tree-sitter tool is functional: parser complete (640/640 rules), emitter produces tree-sitter grammar.js output. Generated grammar has 335+ conflicts requiring iterative resolution. The iterative resolution process (fixing conflicts one at a time with documented rationale) is itself a contribution for the paper.
B.3.5 Automation Results
| Category | % Rules | Handling |
|---|---|---|
| Direct conversion | 38% | Basic syntax maps directly |
| Strip & convert | 55% | Remove annotations, keep structure |
| Best-effort | 6% | Approximate semantic actions |
| Manual review | <1% | Complex disambiguation |
B.3.6 Dual-Path Cross-Validation
Comparing the generated grammar against the hand-written tree-sitter-sysml identifies spec interpretation errors in the hand-written grammar and practical parsing issues in the generated grammar. This cross-validation is a novel contribution.
B.4 INCOSE 2027: SE Benchmark for AI
Working Title: “Toward a Systems Engineering Benchmark for Large Language Models”
Target Venue: INCOSE International Symposium 2027 or Systems Engineering Journal
| Attribute | Value |
|---|---|
| Format | 10-12 page technical paper |
| Target | Q2-Q3 2027 submission |
| Builds On | GVSETS 2026 (MCP server), Grammar Paper (formal methodology) |
B.4.1 Key Thesis
The systems engineering community needs standardized benchmarks to evaluate AI/LLM capabilities on SE tasks, analogous to SWE-bench for software engineering. No standardized benchmark exists for requirements engineering, architecture definition, or V&V — core SE activities. Without benchmarks, progress in AI4SE cannot be measured objectively.
B.4.2 Gap Analysis
| Existing Benchmark | Domain | SE Coverage |
|---|---|---|
| SWE-bench | Software bug fixing | None |
| HumanEval | Code generation | None |
| Humanity’s Last Exam | Expert knowledge | Minimal |
| No existing benchmark | Systems engineering | - |
B.4.3 Proposed Framework
SE task taxonomy aligned with INCOSE processes: requirements elicitation, requirements quality assessment, model completion, test generation from requirements, and requirements-to-design traceability. Each task has deterministic ground truth, measurable evaluation metrics, and supports comparison between baseline AI and MCP-enabled AI conditions.
B.4.4 Relationship to Benchmark Vignettes
The benchmark vignettes defined in Section C.1 (V1-V8) serve as pilot tasks for this paper. The GVSETS paper uses V1, V4, V5 for proof of value; this paper expands to the full set and adds formal evaluation methodology.
B.4.5 Timeline
| Phase | Target | Activities |
|---|---|---|
| Foundation | Current (capstone) | Literature review, MCP implementation, vignette definitions |
| Task Design | Q3 2026 | Define 50-100 tasks, evaluation protocols |
| Pilot Study | Q4 2026 | Run benchmark, collect data |
| SME Validation | Q1 2027 | Expert review of tasks and results |
| Paper Draft | Q2 2027 | Write and internal review |
| Submission | Q3 2027 | Target INCOSE IS 2027 |