thedixitjain ASE Related Work
Position the paper against prior automation, not against a topic. At ASE the question a reviewer holds is "what does this tool/technique do that the existing automated approaches do not?" — so related work is a delta argument, and the strongest deltas are stated against named, runnable prior tools you can (and often must) compare against empirically.
Cover the ASE literature lanes
An automated-SE paper usually sits at the intersection of several lanes; name your position in each that applies:
- Program analysis — static/dynamic analysis, symbolic/concolic execution, abstract interpretation, invariant inference.
- Test generation and testing — search-based, specification-based, fuzzing, regression and flaky-test handling.
- Fault localization, debugging, and repair — spectrum-based localization, automated program repair, patch synthesis.
- Synthesis and recommendation — program/code synthesis, API recommendation, refactoring automation.
- Comprehension and documentation — summarization, code search, clone and change analysis.
- Learning-based / AI4SE — learned code representations, models applied to SE tasks, LLM-based automation — and SE4AI, applying SE automation to ML systems.
Delta-first positioning
Write each related-work cluster as: prior automation X does A; it cannot do B because C; we do B. Then, where B is comparable, compare against X empirically — ASE reviewers expect a head-to-head against the closest runnable baseline, not just a prose distinction.
[Cluster] spectrum-based fault localization
Prior: ranks statements by suspiciousness from pass/fail spectra.
Gap: assumes a fixed test suite; degrades when tests are flaky.
Delta: our technique estimates and corrects for flakiness before ranking.
Evidence of delta: head-to-head on the same faulty subjects, §5.2, Table 3.
Fair head-to-head framing (the ASE credibility test)
- Compare against the strongest, most recent automated baseline you can run, at an equal, documented budget (time, iterations, tuning). An under-configured baseline is the classic soundness objection and a Revision criterion waiting to happen.
- If a baseline tool cannot run on your subjects, say why explicitly and choose the nearest viable alternative — do not quietly omit the obvious competitor.
- Distinguish reimplementation from the original tool, and note version/commit — reviewers who built the baseline will check.
Distinguish ASE from sibling-venue prior work
Because automated-SE ideas appear across venues, position precisely: a testing-theory result may live at ISSTA, a broad empirical study at FSE, a language mechanism at a PL venue. When you cite across venues, make the ASE delta — the automation you add — explicit, so a reviewer does not read your contribution as a reheated sibling-venue result.
Double-anonymous self-citation
- Cite your own prior tools and papers in the third person ("Prior work by [Anonymous] proposed ...", or neutrally "The X tool [12] ...") so the citation does not deanonymize you.
- Do not write "our earlier tool" or link to a repository whose owner is you.
- Keep the count of self-citations honest; a cluster of third-person citations to one hidden group is itself a leak.
Coverage checks
[Recency] at least the last 2-3 years of ASE/ICSE/FSE/ISSTA automation in your subarea cited?
[Closest baseline] is the single most similar runnable tool named AND compared, not just cited?
[Cross-venue] PL/ML/testing neighbors positioned so your automation delta is explicit?
[Anonymity] every self-citation in the third person; no repository ownership revealed?
[Honesty] any obvious competitor conspicuously missing? (a reviewer will notice)
Output format
[Lanes] which automated-SE lanes the paper touches
[Delta table] prior tool -> what it does -> gap -> your delta -> where the head-to-head lives
[Baseline fairness] closest runnable baseline named, version pinned, equal budget?
[Anonymity check] self-citations third-person, no repo owner revealed
Source: brycewang-stanford/Awesome-Journal-Skills → ASE-Skills/skills/ase-related-work/SKILL.md