lzh-phd/topic-feasibility-screener

Topic Feasibility Screener is a Codex skill that helps researchers figure out whether an empirical research idea is actually worth pursuing, starting from an existing CSV or Stata dataset. Instead of asking someone to invent a topic from scratch, it profiles the dataset, brainstorms a broad set of candidate topics from the usable variables, searches real literature for related work, runs a first pass regression to check the data signal, and packages everything into a visual HTML dashboard. The tool is aimed at early stage thesis and paper work in fields like accounting, finance, economics, management, corporate governance, ESG, audit, disclosure, and capital markets research. Its core idea is that a statistically significant result on its own does not make a topic publishable. A good topic needs both theory feasibility, meaning a clear question, related literature, a believable mechanism, and a real gap, and data feasibility, meaning usable variables, enough variation, and a workable sample size. A useful distinction the tool makes is between a topic with no real theoretical footing and one that sits in a genuine frontier gap, where no paper studies the exact relationship yet but nearby literature offers supporting evidence. It sorts the literature it finds into four categories: direct evidence, bridge evidence from related concepts, background evidence, and weak evidence, and treats bridge evidence as valuable rather than simply irrelevant when direct evidence is missing. Running the main script produces an HTML report plus a set of supporting files: score breakdowns, literature diagnostics, a variable profile, and a mandatory self audit file meant to be checked before trusting any of the results. All the major dashboard scores are shown on a 0 to 100 scale, covering empirical feasibility, theory support, theory gap, topic alignment, and an innovation gap score. The README is explicit that the tool's rankings are a screening aid, not proof of a publishable finding, and that final decisions on variables, research design, and literature positioning still require human academic judgment. It installs as a Codex skill and is released under the MIT license.