Elysia Tools
Navigation
Workflow Playbook
Review stock-phrase and repetition findings in an English or Chinese Markdown draft, protect facts and quotations, reconcile revisions and links, and deliver an approved source with a review ledger.
Hubs
This workflow serves editors and technical writers reviewing an existing English or Chinese Markdown draft. Its result is an approved source plus evidence explaining individual style decisions, not a brand-voice redesign or an AI-authorship verdict. Keep the original, record its revision ID and SHA-256, and agree audience, writing context and reviewer. Mark numbers, units, negation, quotations, product names, code and obligations that must not change without approval.
Extract a baseline inventory of link text and destinations. Record extraction options and manually reconcile omitted constructs; an extraction count is not a guarantee of a complete bibliography. The checker runs local rules on this site's backend without a third-party AI call. That is not browser-only privacy. Obtain permission before submission and protect downloaded reports, which contain the original.
Phrase lexicons cover English and Chinese only, even when the interface is translated into other languages. Automatic mode combines those lexicons; sentence grouping by script does not identify a language. The input ceiling is 30,000 UTF-16 code units. For longer drafts, prepare mapped sections without breaking code fences or citations and declare unreviewed ranges. Cross-section repetition is not automatically reconciled.
Record context, lexicon selection, Markdown exclusions, allowlist and report engine/lexicon version. Exclusions are heuristic: code, blockquotes and URLs still need independent preservation checks. A literal allowlist hides matching phrases rather than proving they are appropriate; record why each exception is necessary, not merely why it lowers counts.
Save the JSON and review Markdown. Repetition/rhythm uses at most 1,000 sentences, findings are capped at 300 and repeated-phrase groups at 25; inspect truncation and sampled evidence. The chart shows at most 80 sentences per group, not the full review. Fewer than six sentences in a script group leaves rhythm insufficient; uniform lengths and the CV statistic are not an authorship probability or an editing target.
Maintain your own ledger linking source revision/hash, frozen report, rule, quoted evidence and location to revise, retain or pending, with reason and reviewer. The exported review is not a completed decision log or rewritten article. Finding IDs and positions can change after editing; rerun on the new source and reconcile with context rather than matching only a numbered finding.
For example, “Moreover, the pilot covered three teams.” may become “The pilot covered three teams.” if the transition adds nothing. Keep “three” and the pilot scope. A separate claim that all customers benefit needs supporting evidence and an owner's decision, not a smoother synonym. Protect quotations, uncertainty and words such as may, must and not.
The text diff is a line-level aid, not semantic verification. Inspect actual changed lines and protected spans manually, especially after paragraph reflow. Record justified retained repetition; adding synonyms or changing sentence lengths solely to eliminate alerts can make the draft worse.
Run lint before accepting automatic formatting. Its auto-fix is enabled by default, so disable it for a diagnostic baseline or use a copy. Recheck the resulting diff and rendering: trailing spaces may represent a deliberate Markdown hard break, and list spacing can affect layout. Approve exceptions for code, quotations and house style rather than applying unrelated typography rules blindly.
Re-extract links and compare required destinations and labels with the original. The link checker checks syntax, reference definitions and local heading anchors without network requests. A clean result proves neither external availability nor that a source supports a claim. Heading slug conventions and relative asset paths vary by renderer; test actual navigation and images in the intended publishing context. Verify sources separately when the editorial task requires it.
Preview the exact approved Markdown. If a fixed review artifact is useful, export a PDF with recorded theme and page settings and inspect fonts, tables, code overflow, quotations, links and page breaks. HTML preview and PDF export can render differently; neither substitutes for the target site's final check or the editable source.
Archive original and final Markdown, source inventories, initial and final style reports, a diff, the decision ledger and any reviewed PDF. Record unresolved items, coverage and approver, then hash the actual local deliverables. Hashes identify files rather than certify meaning. Use the related brand-voice workflow for positioning and tone, documentation publishing for release, and localization review for translation. This page stops at traceable style revision, not fact certification, automatic rewriting or detector evasion.
Workflow playbook
Hash the original and extract its Markdown links as a baseline inventory. Record revision, audience, required quotations and facts, and any links the extractor or its chosen options omit.
Run the checker with recorded lexicon, context, Markdown exclusions and literal allowlist. Save JSON and review Markdown exports and inspect scope and truncation warnings before treating the report as complete.
Maintain a separate revise/retain/pending ledger, edit a copy and compare original and revised text. Review every material change to facts, qualifications and quotations; rerun on the current text without treating fewer findings as approval.
Lint the revised Markdown, review any auto-fix on a copy and statically check references and anchors. Compare extracted destinations with the baseline; network availability and source truth require separate verification.
Change a redundant transition only when meaning survives. Keep justified terminology, deliberate repetition and accurate quotations; record a reason rather than forcing the finding count to zero.
Inspect excluded ranges, short-sample rhythm and truncation warnings. An unsupported language or unreviewed section remains outside scope; seven translated interfaces do not provide seven-language lexicons.
Route unsupported claims to a source owner. The linter and line diff do not verify facts or semantic equivalence; numbers, negation, units and obligations require human comparison.
Inspect the actual rendered Markdown and, if needed, a PDF review copy. Save original, final source, reports and ledger with reviewer approval and SHA-256; do not substitute a PDF for editable Markdown.