--- name: brycewang-stanford/cav-topic-selection source: https://app.decimal.ai/s/brycewang-stanford-cav-topic-selection@1/SKILL.md source_sha256: 602120e16556 --- # CAV Topic Selection Decide the venue and the category before drafting. CAV — the International Conference on Computer Aided Verification — is the **flagship** venue for computer-aided formal analysis of hardware and software systems: model checking, SMT and theorem proving, program analysis and synthesis, and their tools. Its papers are Springer LNCS chapters read by verification researchers, so reviewers reward a **durable verification contribution with a stated guarantee**, not a systems demo or an ML result with a verification label attached. ## Two decisions, not one At CAV you choose both a **venue** (CAV vs. its siblings) and a **category** (Regular vs. Tool vs. Application vs. Industrial). Get the venue right first, then the category — a strong tool filed as a Regular Paper, or a technique squeezed into a 10-page tool paper, wastes the fit. ## Sibling-venue routing table | Signal in your project | Better home | Why | |---|---|---| | A general verification technique/algorithm with a soundness/completeness result, mature enough for the flagship | **CAV** | Flagship scope; the Regular-Paper archetype | | Emphasis on tools, algorithms, and their construction/analysis, or you want the ETAPS calendar | **TACAS** | Tools and Algorithms for the Construction and Analysis of Systems; overlaps heavily but is a distinct venue | | Hardware-oriented or applied model checking, or a formal-methods-in-design focus | **FMCAD** | Formal Methods in Computer-Aided Design | | Verification, model checking, and abstract interpretation with a foundations flavor, often earlier-stage | **VMCAI** | Verification, Model Checking, and Abstract Interpretation | | Automated/interactive theorem proving or logic-focused | **IJCAR / LPAR / ITP / CADE** | Reasoning and proof communities | | The core is a programming-language semantics, type system, or PL analysis | **POPL / PLDI / OOPSLA** | PL venues; verification is a means, not the contribution | | The study is too long or too deep for the page limit | **FMSD / JAR / TOCL / STTT** | Journals with no conference page ceiling | CAV and TACAS overlap the most; the honest tie-breakers are the **calendar** (CAV is annual in mid-year, often under FLoC or standalone; TACAS runs at ETAPS in spring) and **community pull** for your subarea. A strong paper is publishable at either — route to the nearer honest fit. ## Contribution shapes CAV rewards - **Technique / algorithm with a guarantee** — a new decision procedure, model-checking algorithm, abstraction, invariant-synthesis method, or proof technique, with a stated soundness/completeness property and evidence it scales past prior methods (the CEGAR / interpolation lineage). - **Tool paper** — a usable, downloadable verification tool (solver, model checker, prover, analyzer) evaluated on standard benchmarks, with the engineering foregrounded (the CVC4 / Marabou lineage). - **New application domain** — bringing a hard real problem into verification's scope with a real technique (the neural-network-verification lineage). - **Application / case study / industrial experience** — applying verification to a concrete system and reporting what worked, what did not, and what generalizes. ## The re-label and swap tests Two quick tests sharpen a borderline verdict: - **Guarantee test:** does your contribution come with a formal property (soundness, completeness, an equisatisfiability or refinement claim)? If the "result" is only a benchmark score with no guarantee, it may be a tool note (TACAS) or a heuristic paper, not a CAV Regular Paper. - **Re-label test:** could this paper be submitted to FMCAD or VMCAI unchanged and read as native there? If its heart is hardware-design methodology or early-stage abstract interpretation, route accordingly; CAV rewards the general, flagship framing. ## Category-selection cues (once CAV is chosen) ```text [Regular] a technique/algorithm + proof + benchmark evaluation -> 18 pages, anonymized [Short Tool] a downloadable tool; the contribution is the usable system -> 10 pages, NOT anonymized [Short App] verification applied to a specific problem, technique-light -> 10 pages, anonymized [Industrial] a real/industrial deployment experience or case study -> 10 pages, NOT anonymized ``` If you have both a technique and a tool, the usual move is a Regular Paper that describes the technique with the tool as its evaluation vehicle — reserve the Tool Paper for when the *system itself* is the contribution. ## Cheap reconnaissance before committing ```text [Scope] scan the last two CAV programs (dblp, i-cav.org) for your subarea -> 3+ recent papers = a reviewer pool exists; 0 = opening or mismatch [Benchmarks] is there a standard benchmark set (SV-COMP, SMT-COMP, HWMCC, VNN-COMP) reviewers expect? -> yes and you did not use it => reframe or add it before submitting [Calendar] compare the next CAV deadline with TACAS/FMCAD/VMCAI dates -> route to the nearest honest fit rather than idling a cycle ``` ## Decision procedure ```text [Audience] who acts differently if the claim holds? -> verification-tool builders/users/theorists? [Claim type] technique-with-guarantee / tool / application / industrial experience [CAV vs sibling] both fit? -> choose by calendar, community pull, and hardware/PL/logic tilt [Category] technique+proof -> Regular; usable system -> Tool; applied -> Application/Industrial [Verdict] CAV / sibling venue / journal, with a one-line reason ``` Run this before the writing skills; a wrong venue or category decision wastes every later step. When the verdict is CAV, continue with `cav-workflow` for the calendar and `cav-writing-style` for the paper shape.