writeArticleFile now writes to a potentially suffixed filename via resolveUniqueFilePath(), but it still only returns a boolean. Callers like writeSingleArticle() (same module) record filename into stats.articles, so a deduplicated write will cause metadata/stats to reference a file that was never created. Consider returning the actual written filename/path (or exposing it via an out param) and updating stats/logging to use that value.

resolveUniqueAnalysisDir(preferredDir) can produce a suffixed directory like analysis/{date}/{slug}-2, but the article transparency links are built from {date} + {slug} only. On repeated same-day runs, newly generated articles will link to the original analysis directory instead of the run-specific suffixed one, breaking provenance. Consider passing the resolved analysis directory name (from AnalysisContext.outputDir or the manifest) into generateArticleHTML() / renderAnalysisTransparencySection() so links match the actual output path.

This section hard-codes analysis links to analysis/{date}/{slug}. With suffix deduplication in the analysis stage (e.g. analysis/{date}/{slug}-2), repeated workflow runs will produce articles whose Analysis & Transparency links point to the wrong (earlier) analysis artifacts. The template likely needs an additional parameter (e.g. analysisSlug / analysisDirName) so it can link to the resolved analysis output directory for the specific run.

The new per-file analysis links use English-only link text (e.g. “Significance Assessment”, “Actor Mapping”) even when generating non-English articles (<article lang="...">). Since the surrounding headings/labels are localized, this is a regression for multilingual UX and screen readers. Consider localizing these link titles (or at least reusing existing localized labels where they exist) so the section language matches the document language.

resolveUniqueAnalysisDir(preferredDir) can return a suffixed path (e.g. analysis/{date}/{slug}-2), but article generation and linking logic elsewhere appears to still key off {date} + {slug}. That means repeated same-day runs can generate new analysis in a suffixed directory while produced articles still point to / consume the original directory, breaking run provenance. Consider propagating the resolved output directory (or its basename) through the pipeline so consumers link to the correct run folder.

These Analysis & Transparency links are constructed as analysis/{date}/{slug}. With suffix-based deduplication in the analysis stage (e.g. analysis/{date}/{slug}-2), articles generated on repeated runs will link to the wrong (earlier) analysis artifacts. Consider passing a run-specific analysis directory name into the template so it can link to the resolved analysis output path for that run.

The new per-file analysis links use English-only link text (e.g. “Significance Assessment”, “Actor Mapping”) regardless of the article language (lang). Since surrounding headings are localized and the HTML document sets lang, consider localizing these link titles (or reusing existing localized labels) to avoid mixed-language navigation for screen readers and non-English readers.

dedupSuffix extraction only supports numeric -(\d+) suffixes. If resolveUniqueAnalysisDir() falls back to the UUID suffix (or any future non-numeric suffix), article slugs won’t include the suffix, so repeated runs can still overwrite news/{date}-{type}-{lang}.html and the article’s canonical/og:url won’t line up with the deduplicated analysis directory.

Consider propagating the full suffix from the resolved analysis directory (numeric or UUID) and appending that to each strategy.type, instead of restricting to digits only.

runAnalysisStage() calls resolveUniqueAnalysisDir(preferredDir) before validating hasSubstantiveData(fetchedData). Since resolveUniqueAnalysisDir may create a suffixed directory immediately, a run that later aborts (e.g. requireData=true with empty/invalid data, or another preflight error) can leave behind an empty claimed directory (e.g. .../breaking-2) that forces subsequent runs to skip to -3, etc. Consider moving the substantive-data validation earlier (before claiming/creating any directories), or cleaning up the claimed directory on early failure.

deriveTypeSlug() returns the full analysisDir when it starts with ${strategyType}-, which breaks when analysisDir is a multi-type slug produced by deriveArticleTypeSlug (e.g. breaking-week-ahead-2): the “breaking” article would incorrectly get typeSlug breaking-week-ahead-2. It also does not implement the PR’s stated behavior of extracting the dedup suffix (e.g. -2) and applying it per strategy type. Consider parsing the suffix from analysisDir (numeric/UUID) and appending it to each strategyType, instead of substituting analysisDir wholesale.

news-enhanced.ts passes a single analysisDir (basename of analysisCtx.outputDir) to generateArticleForStrategy() for every requested article type. When multiple articleTypes are generated in one invocation, the analysis output directory slug is derived from all types (deriveArticleTypeSlug), so analysisDir will not match each individual strategy’s analysis folder. This will make Analysis & Transparency links (and potentially the deduped type slug logic) point to the wrong directory for some articles. A safer approach is to compute a per-strategy analysisDir (or just a suffix) and pass that per strategy instead of reusing the run-level basename for all strategies.

PR description says “Any existing directory … is treated as occupied” for slug-level dedup, but resolveUniqueAnalysisDir() currently treats a directory as occupied only when manifest.json exists (and even comments that a directory without manifest is considered available). If the intended contract is “any existing dir is occupied”, the implementation and JSDoc/comment here should be updated to match (or the PR description adjusted).

The new dedup-related behavior (deriveTypeSlug + propagation of analysisDir into generateArticleHTML for canonical URLs / transparency links) is not covered by unit tests. There are existing generate-stage pipeline tests, but none assert that a suffixed analysis run (e.g. breaking-2) results in (a) the expected article slug/filename pattern and (b) transparency links pointing at the suffixed analysis directory. Adding focused Vitest cases for numeric and UUID suffixes (and for multi-article-type slugs) would help prevent regressions.