Research question: Do the 12 campaign articles named in the publication-quality test meet a reproducible minimum standard for substantive length, independent structure, citation practice, and removal of templated filler?
Executive result
Yes, for the stated mechanical rubric. All 12 expected rewritten files are present. Every article reaches the 600-substantive-word floor, has at least five H2 headings, avoids the prohibited filler expressions, and contributes a distinct ordered H2 sequence. Across all 66 pairs, the highest five-word-shingle Jaccard score is 0.005, below the unchanged 0.20 ceiling.
None of the practical bodies contains an external Markdown citation. That is reported separately rather than treated as a failure because these articles avoid external numerical, legal, market, or performance claims; property- and client-specific facts are left to controlled evidence and accountable approval. These are repository observations, not estimates of reader satisfaction or search performance.
Scope and denominator
The denominator is exactly 12 article files: the fixed slug list declared in lib/__tests__/campaign-content-quality.test.ts. The audit unit is one MDX article body. Frontmatter was available but excluded from body word and similarity calculations. The review did not sample other articles, research reports, images, rendered pages, analytics, leads, transactions, or search results.
The observations describe the files present in this repository snapshot at review time. “12/12” therefore means 12 of these 12 named campaign files, not 12 of every article published by Real Estate Luxury. “External citation” means an HTTP or HTTPS destination in a Markdown link in the article body. It does not mean an unlinked source name, a canonical URL in frontmatter, or evidence held outside the file.
Audit method and rubric
The measurement reproduces the test logic rather than applying a subjective editorial score. For each expected slug, the audit checked that the corresponding .mdx file existed and then evaluated its Markdown body as follows:
- Remove Markdown heading lines and replace each Markdown link with its visible anchor text. Convert the remaining body to lowercase and count tokens matching letters or numbers with optional internal apostrophes or hyphens. A passing article needs at least 600 such substantive words.
- Extract every line beginning with
##, normalize the heading text to lowercase, and preserve order. Each article needs at least five H2 headings, and the 12 complete H2 sequences must all be distinct. - Build sets of consecutive five-word shingles from the substantive tokens. Calculate Jaccard similarity for each of the 66 possible article pairs. The highest pairwise score must be below 0.2.
- Search the body, case-insensitively, for the prohibited expressions “Campaign item,” “Define the reader question,” and “Keep claims traceable.” A passing file contains none of them.
- Count external Markdown-link destinations separately as an editorial evidence measure. The current automated article test does not require a citation, so citation status is reported rather than silently folded into the code-test result.
The full campaign acceptance decision is conjunctive: inventory, individual length, minimum heading count, removal of prohibited filler, corpus-wide heading independence, and corpus-wide similarity must all pass. One successful check cannot offset a failed check. For example, seven H2 headings satisfy the heading-count floor but do not cure a repeated outline.
Measured audit results
| Metric | Observed result | Denominator or unit | Acceptance interpretation |
|---|---|---|---|
| Expected article files found | 12/12 | 12 named slugs | Inventory passes. |
| Articles with at least 600 substantive words | 12/12 | 12 article bodies | Length floor passes. |
| Articles with at least five H2 headings | 12/12 | 12 article bodies | Heading-count floor passes. |
| Distinct H2 sequences | 12/12 | 12 ordered sequences | Structure-independence passes. |
| Articles containing an external Markdown citation | 0/12 | 12 article bodies | No external factual claims were added. |
| Articles containing prohibited filler | 0/12 | 12 article bodies | Filler-removal passes. |
| Highest pairwise five-word-shingle similarity | 0.005 | 66 article pairs | Originality passes below 0.20. |
| Articles passing the complete campaign rubric | 12/12 | 12 article bodies | The deterministic contract passes. |
Body counts range from 750 to 899 substantive words. The worst pair is the daily editorial brief and editorial correction record. Naming it preserves reproducibility; the low score does not alone prove originality or authorship.
What the results mean
The current corpus clears the mechanical publication floor with topic-specific architectures. The maintenance brief centers an evidence packet and authorization path; the private-showing article uses invitation and exposure checks; the correction record follows discovery, impact, correction, and propagation. That variation agrees with the measured heading and shingle results.
Zero external links does not prove every sentence true, but it is appropriate here because the articles present first-party workflow guidance and avoid public-standard, legal, market, and performance claims. A future externally verifiable proposition should add a directly relevant source mapped to that claim.
Similarity is a corpus control, not a diagnosis of plagiarism or intent. Common domain terms can overlap, which is why the rule permits a score below 0.20. The observed 0.005 provides substantial margin without replacing direct editorial inspection.
Data sources and references
The external references below define relevant publishing practices; they are not housing-market or economic proxies.
- Google’s people-first content guidance asks whether content serves an existing or intended audience and demonstrates first-hand expertise. It supports an audience-and-purpose review, but it does not promise a ranking.
- Google’s SEO Starter Guide recommends well-organized, unique, useful content and descriptive headings. Those practices can improve clarity and crawl understanding; they do not guarantee visibility or traffic.
- Google’s Article structured-data documentation explains that Article markup can help Google understand an article’s title, images, dates, and author. Eligibility or valid markup does not guarantee a search feature.
- Google’s byline-date documentation recommends a prominent user-visible date and describes
datePublishedanddateModifiedfields. It supports consistent date labeling, not a freshness or ranking claim. - W3C WAI’s writing accessibility tips call for informative unique page titles, meaningful headings, and descriptive link text. These are checkable editorial controls that benefit navigation and comprehension.
- W3C WAI’s heading tutorial explains that headings communicate page organization and should be nested by rank. A heading audit should therefore inspect hierarchy and meaning, not merely count headings.
- W3C WAI’s alt decision tree bases alternative-text treatment on an image’s purpose, including different treatment for informative, complex, redundant, and decorative images. A generic filename check cannot replace that contextual decision.
- GOV.UK’s writing standards guidance frames content design around clear, helpful content and identified user needs. It supports planning around a concrete task rather than filling a generic outline.
- Digital.gov’s plain-language guide series describes plain language as content a specific audience can understand and points to designing and testing for that audience. It does not reduce quality to short sentences alone.
- Schema.org’s Article vocabulary defines the Article type and its available properties. It can supply a common metadata vocabulary, while implementation and validation remain separate editorial and technical checks.
Together, these sources support a layered review: useful topic-specific substance, understandable writing, navigable structure, purposeful image alternatives, consistent dates, and accurate article metadata. They do not establish that any particular luxury real estate claim is true. Property facts still require property-level evidence and accountable approval.
Acceptance readback
The imported campaign was checked in the same paths and fixed denominator. Readback confirms 12/12 files, 12/12 at or above 600 substantive words, 0/12 containing prohibited filler, 12/12 with at least five H2 headings, 12 distinct H2 sequences, and maximum similarity 0.005 across 66 pairs.
Evidence and human review remain separate gates. Future external publishing, accessibility, search, market, property, or legal claims need source-specific support. Titles and links should describe destinations, dates should remain consistent, image alternatives should match image purpose, and a named reviewer should approve factual scope.
Limitations and inference boundary
This is a static repository audit, not a reader study. It does not measure claim-level accuracy, assistive-technology behavior, rendering, mobile performance, expertise, comprehension, leads, conversions, transactions, or rankings. Reachability checks establish only that a cited source responded when reviewed.
The audit establishes that the observed files satisfy the stated mechanical rubric on the review date. It cannot establish why prior defects occurred, whether a page will rank, whether readers will act, whether client information is protected, whether a property statement is legally sufficient, or whether a commercial outcome will improve. No publishing checklist guarantees search placement, privacy, legal compliance, accessibility, factual correctness, or business results.