# ScholarFlow Project #19: Methodology, constraints, and reproducibility ## Purpose ScholarFlow supports **organizing** a research review, not replacing scholarly judgment. A user can catalog records, document screening decisions, inspect the included set and export a transparent review log. ## Inputs and provenance Initial records are synthetic teaching examples, labeled `synthetic: true`; they are not articles, studies, results or verified references. Real bibliographic metadata and evidence must be collected and verified outside this tool. A user-added record has `synthetic: false`, which means only that the user created it, **not that a claim has been validated**. ## Screening decisions The available states are `unscreened`, `include`, `maybe`, `exclude`. Their meaning should be set within the user's review protocol and recorded in free-text notes. This release intentionally does not infer screening decisions from titles or abstracts. ## Evidence matrix Only records explicitly marked `include` appear in the evidence matrix. A five-field **metadata completeness** count tests whether the source, method, findings, limitations and participants fields have values. The synthetic source placeholder is not counted as a valid source field. This count is never a quality grade, bias rating or proof of valid findings. ## Coverage analysis The gap explorer reports frequencies for the `region` and `language` metadata fields among included records and raises flags when fewer than three distinct methodological, regional or linguistic categories appear. A record set of fewer than ten included items is flagged as small. The threshold is an explicit heuristic for prompting review, not a measure of statistical significance or research neglect. All interpretations are restricted to the imported library. ## Storage and reproducibility The application uses browser-local `localStorage` under `scholarflow-project-19-v1`. It makes no API calls or uploads. JSON exports contain all records and reviewer notes, an export timestamp, schema version, and a warning that demo data are synthetic. CSV exports are useful for tabular inspection. Markdown exports are useful for human-readable audit trails. The same dataset and decisions will produce the same summary counts on the same app version. ## Responsible academic use - Verify each paper's existence, DOI, metadata and original claims against the primary source. - Predefine question, eligibility rules and search strategy before conclusions. - Preserve reasons for exclusion and unresolved screening disagreements. - Assess study quality with a **suitable domain-specific validated rubric**, not the completeness count. - Do not convert descriptive coverage counts into claims about the whole field. - Cite publications themselves, never synthetic demo records. - Export backups before browser resets or data imports. ## Non-goals This version is not a PRISMA-compliant search workflow, automated systematic review, external literature search, model-powered summarizer, statistical meta-analysis, or a publication-verification service.