Approach: Nielsen Heuristics
Run Chimera's get-skill tool with arguments {"skill_name": "inventory_from_template", "approach_slug": "nielsen-heuristics"}. Both arguments are required — pass approach_slug as shown. Then follow the workflow embedded in the response.
Jakob Nielsen's 10 Usability Heuristics are the most widely cited rules of thumb for interface usability — distilled by Nielsen in 1994 from a survey of usability problems and updated since. They are deliberately broad principles rather than specific guidelines, intended to be applied with judgment to whatever interface is in front of you.
Applied to web content as a rubric, the heuristics let you score every page on the same ten dimensions: how clearly the page communicates where the user is, whether its language matches how the audience actually thinks, whether users can recover from mistakes, whether the layout is consistent with the rest of the site, and so on.
Each page gets a score for each heuristic plus an overall score, so you can sort an inventory by usability, find your weakest pages or weakest dimensions, and use that to prioritize rework.
Several heuristics are hard to fully automate. Visibility of System Status and Aesthetic and Minimalist Design depend on visual rendering, motion, and feedback. Help Users Recognize, Diagnose, and Recover From Errors depends on what actually happens when something goes wrong — observable only by exercising the interface. Any automated approach will need to either approximate these from indirect signals or surface them for human review.
Error Prevention and Error Recovery do not apply to purely informational pages. Treating those as n/a rather than scoring them low, and skipping n/a values when averaging, keeps content pages from being unfairly penalized for not having forms.
Content-heavy sites tend to score consistently lower on Flexibility and Help & Documentation because the criteria are taken literally. If your site is informational by design, treat low scores on those two heuristics as expected rather than as defects.
For high-stakes evaluations, pair automation with human spot-checks on representative pages.
Chimera scores each page based on its text — rendered HTML stripped to plain text — so heuristics that depend on visual rendering or interactive behavior are inferred from text and structure rather than directly observed. A seeded language-model fieldset returns a 0-3 rating and a written rationale for each heuristic. For H4 Consistency, three XPath patterns (count of H1 tags, count of landmark regions, count of images missing alt text) are also extracted and blended with the language-model rating. The per-heuristic deliverables and the overall average are produced as calculated fields in the extent's flattened table.
The Run Now in Chimera button on this page launches the inventory-from-template skill, which copies the seeded fieldset and patterns into your extent, runs the summarize and extract pipelines, and creates the calculated fields automatically. You can optionally compose the Nielsen fieldset with an org-type baseline (minimal, higher_ed, research) or with E-E-A-T page analysis to capture additional dimensions in the same language-model call.
Supporting
Raw inputs that produced the primary fields above — useful for drill-down.
Helper
Computational plumbing — exists to make the primary formulas work. Usually not interesting on its own.