How the Score is Calculated

Your Cassian Score™ is a weighted composite of up to five layers. Each layer measures a distinct aspect of your store's health. Every time a scan runs, Cassian™ evaluates the crawled pages of your store — using a mix of deterministic checks and AI analysis — and feeds the results into this model.

Which layers are active depends on your plan and store: the free Inspector plan scores Technical Health, Content Quality, and LLM Readiness. Paid plans add Platform Consistency, plus Translation Quality when your store has translations. Inactive layers are excluded and the remaining weights are renormalised to total 100%.

Here's exactly what's inside each layer.

---

Layer 1 — Technical Health (30%)

Technical Health is the heaviest layer, and intentionally so. Technical problems are the most likely to affect your customers directly — broken links, missing metadata, and rendering problems hurt both search visibility and user experience.

What Cassian checks on every crawled page:

• Broken links — Internal links returning 4xx or 5xx errors are the most serious; broken images and media, external broken links, and server errors are also flagged (at lower severities).
• Meta titles — Checks for missing or empty titles, multiple <title> tags on one page, titles under 10 characters, and titles over 60 characters.
• Meta descriptions — Flags pages with no meta description.
• Canonical tags — Flags pages with no canonical link tag. Missing canonicals can cause search engines to index duplicate pages.
• Heading structure — Flags missing H1s, multiple H1s, and heading structures that skip levels (e.g., H1 → H3 with no H2).
• Image alt text — Images without descriptive alt text are flagged. This affects both accessibility compliance and SEO.
• SSL & mixed content — Flags resources (images, scripts) loading over HTTP on an HTTPS page.
• Mobile & language basics — Missing viewport meta tag, missing lang attribute, and hreflang inconsistencies.
• Performance signals — Images without declared dimensions (a layout-shift risk) and render-blocking scripts observed during the crawl. For detailed daily Lighthouse scores, see the PageSpeed page — those are monitored separately and don't feed this layer.

Technical Health combines instant, rules-based checks with AI analysis of each page. The deterministic checks are consistent between scans; the AI checks add findings that rules can't catch.

---

Layer 2 — Content Quality (25%)

Cassian AI™ reads the actual text on your pages — not just the structure — and evaluates it for quality, completeness, and clarity. This layer is about whether the words on your store are doing their job.

What Cassian checks:

• Spelling and grammar — Flags errors in product titles, product descriptions, collection page copy, and blog posts. Even one or two errors on a product page can undermine customer trust.
• Thin content — Product pages with fewer than 50 words of content are flagged at high severity; other pages with fewer than 30 words are flagged at medium severity. Thin descriptions hurt both conversion rates and search rankings.
• Duplicate content — Product descriptions that are identical or near-identical across multiple pages. Common when product variants share descriptions or when catalogue content is copied from a supplier.
• Missing product images — Product pages with no product imagery are flagged.
• Readability and clarity — Flags content that reads as unusually complex or uses jargon inconsistent with an e-commerce context. Cassian AI evaluates this in the context of your product category.

---

Layer 3 — Translation Quality (20%)

Translation Quality is only active when your store has translated content. If your store is monolingual, this layer is excluded and the remaining layers are reweighted to total 100%.

What Cassian AI checks across translated content:

• Machine translation artifacts — Phrases and sentence structures that appear to be unedited machine translation output. These are often grammatically correct but stylistically off, or use terminology that doesn't match industry norms in the target language.
• Missing translations — Pages, product descriptions, or metafields where translated content is absent entirely. A customer browsing in French who lands on an untranslated product page is likely to leave.
• Variant translation completeness — Product variants (size, colour, material, etc.) that haven't been fully translated. Incomplete variant translations are one of the most common translation gaps.
• RTL rendering issues — For right-to-left languages such as Arabic, Hebrew, and Persian — flags content that doesn't display correctly in RTL layout contexts.
• Metafield translation accuracy — Compares translated metafields against the source language content to verify completeness and consistency.

---

Layer 4 — Platform Consistency (15%)

Platform Consistency requires a paid plan with a connected platform (Shopify, WooCommerce, or BigCommerce). Cassian pulls data from your platform backend and compares it to what actually renders in your storefront. This catches a category of problem that no other tool finds automatically.

What Cassian checks:

• Product data mismatches — Price, title, or availability displayed on the storefront differs from the data in your platform admin. This can happen due to theme bugs, caching, or CDN issues.
• Image completeness — Product variants missing assigned images. Customers selecting a colour or size and seeing no image is a common conversion killer.
• Translation metafield accuracy — Translated content stored in your platform vs what actually renders in the translated storefront.
• Availability inconsistencies — Products or variants showing as purchasable when they're out of stock, or showing as unavailable when inventory exists.

---

Layer 5 — LLM Readiness (10%)

AI-powered shopping assistants — embedded in search engines, browsers, and retail platforms — are increasingly influencing purchase decisions before a customer ever visits a store. LLM Readiness measures how well your store is positioned for this shift.

What Cassian checks:

• Structured data completeness — The presence and completeness of Product, Organisation, and Breadcrumb schema on your pages. Rich structured data is how AI systems extract product information reliably. Missing Product schema on a product page is one of the highest-severity findings in this layer.
• Robots meta tag — Flags pages with no robots meta tag; it helps search engines and AI crawlers understand indexing preferences.
• Content extractability — Whether product information can be cleanly extracted from your pages. Heavy JavaScript rendering, content behind interactions, and unusual DOM structures can reduce extractability.
• Brand consistency — Consistent brand naming, product naming conventions, and description patterns across your catalogue. Inconsistency makes it harder for AI systems to build a coherent picture of your store.
• Answer-ready formatting — Product descriptions written in complete sentences, rather than pure bullet-point attribute lists, are significantly more usable by AI assistants answering customer queries.

Cassian's separate feeds feature can generate an llms.txt file for your store — but no scored check looks for one, so its presence or absence doesn't move this layer.

---

How the weights work

Whenever a layer is inactive — Translation Quality on a monolingual store, Platform Consistency without a paid plan and connector, or both on the free Inspector plan — that layer is excluded and the remaining weights are renormalised proportionally so the total still equals 100%. The exact reweighted percentages are calculated automatically — you don't need to configure anything.

Coverage, confidence, and normalisation

Three mechanics keep the score fair and stable:

• Coverage factor — The composite score is scaled by how much of the planned page set the scan actually covered: the multiplier is 0.8 + 0.2 × coverage, so a fully completed scan scores at 100% and even a barely-started one never loses more than 20% to coverage. If page discovery itself failed, the scan can't know what it missed, so the factor floors at 0.8.
• Per-page normalisation — Deductions in each layer are normalised against the number of pages that layer actually evaluated (with a minimum denominator of 10), so a small store with one broken link isn't punished more than a large store with one broken link.
• Confidence gating — Every AI finding carries a confidence estimate. Findings below 40% confidence deduct nothing from the score (they're shown for information only); findings between 40–70% apply a sharply reduced deduction; findings at 70%+ apply near-full impact.

---

Common questions

Why is Technical Health weighted so heavily? Technical problems are the most likely to directly impact your store's search visibility and customer experience. A store with broken internal links and missing meta tags is genuinely less healthy than a store with thin descriptions. The weighting reflects the real-world impact of each layer.

What if my store is only in English? Then Translation Quality is simply excluded from your score — it isn't scored at 100, it's removed from the calculation entirely, and its 20% weight is redistributed across the other active layers.

Can the score be 100? Yes. A score of 100 means Cassian found zero issues across all active layers during that scan. It's genuinely achievable with a well-maintained store — though maintaining it requires staying on top of new issues as your store content changes.

What's the difference between Critical, High, Medium, Low, and Info severity? Severity determines how much an issue deducts before normalisation: Critical deducts 15 points, High 8, Medium 4, Low 2, and Info 0 (advisory only). Those raw deductions are then adjusted by confidence and normalised per page as described above. The Issues page always shows severity alongside each finding.

Why did my score drop overnight when I didn't touch anything? Cassian scans your live store. If any page content, product data, or linked resources changed — including third-party scripts, partner feeds, or inventory status — those changes may have introduced new issues. Check the Issues page for newly detected findings to trace what changed.