Content Grader versions

Audit

• Added an Internal Link Graph panel. It reads every anchor in the pasted markup and reports internal links, external links, unique targets, empty hrefs, in-page anchors, generic anchor text, image-only links, links that open a new tab without rel="noopener", nofollow on internal links, and one anchor phrase pointing at several destinations.
• The panel can also check orphans. Paste one URL per line from your sitemap or a crawl export and it lists the pages that no pasted link points at. A target that is missing from your list is reported as not in your list, because that is not the same as a broken link.
• The panel is an audit only. It does not add a module, change a weight, or move the 0-100 score.

Fixes

• Fixed a crash on short input. When the readability module returned no metrics, saving the run threw an error and a blocking alert appeared. The save is now null-safe and errors use the in-page toast.
• Hardened the link graph against the URL shapes that producers paste. A list entry without a scheme, a protocol-relative link, a trailing slash, a query string, a fragment, an uppercase path and a scheme-less domain in an anchor all resolve to the same page key. Relative links resolve against the canonical when the page provides one. Non-web schemes such as mailto, tel and sms are counted as non-page links instead of internal ones, and an in-page anchor can no longer be reported as an anchor phrase pointing at two destinations.

Tests

• Expanded the test harness from fourteen fixtures to twenty-four. The new fixtures cover a link profile with three defects, clean internal linking, orphan detection against a pasted URL list, and seven URL-shape cases: a scheme-less domain, a URL list without a scheme, protocol-relative links, equivalent URL variants, non-web schemes, fragment anchors beside page links, and relative links resolved against a canonical.

Scoring

• Rebuilt Advanced Keyword Intelligence. It now counts exact phrase matches, measures density against real word counts, and scores placement across title, description, H1, H2, the first 100 words, the slug and image alt text.
• Keyword spread is scored across the opening, middle and closing thirds. A phrase that appears only in the intro loses points even when the total count looks healthy.
• The keyword module and the keyword panel now run the same calculation, so the score and the panel can no longer disagree.
• Rebuilt Cross-Platform Optimization. It now scores mobile reading (long paragraphs, unbroken strings, wide tables, long code lines), accessibility (image alt text, H1 count, heading order, generic link text, all-caps runs), link hygiene, and media weight.
• Both modules previously returned no findings because their patterns were double-escaped. The bug is fixed and covered by fixtures.

Tests

• Expanded the test harness from ten fixtures to fourteen.
• New fixtures cover correct keyword placement, a focus keyword missing from the body, markup that breaks on mobile and for assistive tech, and clean accessible markup. The keyword fixtures set the focus keyword, title, description and slug before scoring.

Scoring

• Replaced the flat average with six weighted buckets: answer clarity 25, structure 20, evidence 20, readability 15, entity coverage 10, hygiene 10.
• Added applicability rules. Modules that need page markup, a focus keyword, media, or enough text are marked not applicable and removed from the denominator instead of scoring zero.
• Inputs under 150 words are provisional. The grade letter is hidden because a short snippet cannot support a page-level grade.
• Added a priority panel with the three fixes that would move the score most, each with an estimated point gain.
• Fixed the sentiment module producing a non-numeric score on short inputs.
• Fixed the readability module allowing scores above 100 and disagreeing with its own formula panel.

Answer quality

• Added the answer capsule detector, with the excerpt that an AI engine is most likely to quote.
• Added the entity and claim ledger. Figures without a nearby source marker and absolute claims are listed without changing the score.
• Section opening checks now compare the first paragraph with the heading instead of scanning the whole section.

Exact page audit

• Added an unscored audit for title, description, canonical, robots, Open Graph, Twitter card, viewport, hreflang, JSON-LD syntax, images, links, conversion, trust, pricing, and FAQ schema alignment.
• Head checks now run only when page source is present. Without source, they report not run instead of a false failure.
• Body-only pastes run body checks without false head failures, and the SEO fields can supply title and description.
• Fixed FAQ detection on pages that mark questions with a trailing plus, and on homepages that use a heading outside the FAQ block.

Comparison and transparency

• Added pairwise comparison with phrase overlap and a plain verdict.
• Published the methodology page with the weights, grade bands, module list, and limits.
• Published this changelog so every score claim can be traced to a version.
• Added the fixture-based test harness used to catch regressions.

First documented baseline

• 20 content modules: information gain, citation probability, answer engine optimisation, E-E-A-T, semantic entities, topical authority, skimmability, schema, freshness, originality, multimodal signals, SEO structure, entity salience, GEO, AI detection, readability, keyword intelligence, sentiment, entity extraction, and cross-platform structure.
• Browser-only analysis. No content is uploaded. URL imports send only the URL to a public reader service.
• Exports to HTML, Markdown, JSON, CSV, and PDF.