<!-- Generated from models/_canonical by scripts/build-model-variants.js. Edit the canonical source, not this file. Behavioural profile for Grok: scripts/model-profiles.json -->

#Non-negotiable

The constraints hoisted below override anything later in this document. Read them first; the rest is rationale.

---

#Non-negotiable constraints

These override anything later in this document.

• Never mark up data the page does not display. It is a policy violation and attracts a manual penalty. Validate with the Rich Results Test before shipping.

---

#Purpose

Rules for making a web application discoverable. Scope is technical SEO — what engineering controls. Content strategy and keyword research are not covered.

The order of impact, which is not the order teams usually work in:

1. Can crawlers reach and render the content at all?
2. Is each page's canonical identity unambiguous?
3. Is it fast on mobile?
4. Is it marked up so results can be rich?

---

#Rendering: content must exist without JavaScript

A client-rendered page ships an empty <div id="root">. Google will usually render it eventually; other crawlers, social preview bots and AI crawlers often will not, and rendering is deferred and unreliable.

Verify with curl rather than devtools, which shows the hydrated DOM:

bash
curl -sL https://example.com/products/widget | grep -c "<h1"

If the content is not in that output, assume it may not be indexed. → Frontend/server-components

---

#One canonical URL per page

Duplicate URLs split ranking signals between copies and can cause the wrong one to be indexed.

html
<link rel="canonical" href="https://example.com/products/widget" />

Sources of accidental duplication, all common:

• http and https; www and apex — pick one and 301 the rest.
• Trailing slash inconsistency (/about and /about/).
• Tracking parameters (?utm_source=…) creating infinite variants.
• Uppercase and lowercase paths on a case-sensitive server.
• Pagination and filtered views of the same collection.

Rules:

• A self-referencing canonical on every indexable page, absolute, matching exactly the URL you want indexed.
• Filtered and sorted views (?status=paid&sort=-createdAt) canonicalise to the unfiltered collection, or are noindex.
• Never change a URL without a 301. → Frontend/routing
• hreflang for localised variants, and every variant must point back at all the others including itself.

---

#Metadata that actually renders

html
<title>Widget — Acme</title>                        <!-- ≤ 60 chars, unique per page -->
<meta name="description" content="…" />             <!-- ≤ 155 chars, unique -->
<meta property="og:title" content="Widget" />
<meta property="og:image" content="https://example.com/og/widget.png" />  <!-- absolute -->
<meta name="twitter:card" content="summary_large_image" />

• Generate metadata server-side. A <title> set in useEffect is not seen by most crawlers or by social preview bots at all.
• Unique per page. Duplicated titles and descriptions across a catalogue are one of the most common technical SEO faults.
• Open Graph images must be absolute URLs — relative ones silently fail in every social preview.
• robots meta only where you mean it. A stray noindex shipped to production will remove a site from search results, and it is a genuinely recurring incident. → Frontend/metadata

---

#Structured data

json
{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Widget",
  "offers": { "@type": "Offer", "price": "24.99", "priceCurrency": "EUR",
              "availability": "https://schema.org/InStock" }
}

JSON-LD in a <script type="application/ld+json">, generated from the same data the page renders. Types worth implementing: Product, Article, FAQPage, BreadcrumbList, Organization, Event.

Never mark up data the page does not display. It is a policy violation and attracts a manual penalty. Validate with the Rich Results Test before shipping.

---

#Crawlability

makefile
# robots.txt
User-agent: *
Disallow: /api/
Disallow: /admin/
Sitemap: https://example.com/sitemap.xml

• robots.txt blocks crawling, not indexing. A blocked URL can still be indexed from external links — and because it cannot be crawled, the noindex tag on it is never seen. To keep a page out of the index, allow crawling and use noindex.
• Generate the sitemap from the route table, include lastmod, and split above 50,000 URLs.
• Never block CSS or JavaScript in robots.txt — crawlers need them to render.
• Fix crawl waste: infinite calendars, faceted-filter URL explosion, and session ids in URLs consume crawl budget that should go to real pages.
• Return real status codes: 404 for missing, 410 for permanently gone, 301 for moved. A "not found" page returning 200 is a soft 404 and gets indexed.

---

#Speed and mobile

Core Web Vitals are a ranking signal, measured from field data at p75, not from a Lighthouse run.

Indexing is mobile-first: the mobile rendering is what is evaluated. A desktop page that is fast and a mobile page that is not means the slow one counts. → Frontend/performance

---

#Anti-patterns

---

#Checklist

• Verify: Public content is present in the server response, verified with curl
• Verify: Every indexable page has a unique, server-rendered title and description
• Verify: A self-referencing absolute canonical is on every indexable page
• Verify: One host and one trailing-slash convention, with 301s enforcing them
• Verify: Filtered and sorted views canonicalise or are noindex
• Verify: URL changes are always accompanied by permanent redirects
• Verify: Open Graph and Twitter tags use absolute image URLs
• Verify: No accidental noindex reaches production; production is verified
• Verify: Structured data is JSON-LD, matches visible content, and validates
• Verify: robots.txt allows CSS and JavaScript and links the sitemap
• Verify: The sitemap is generated from routes, with lastmod and size splitting
• Verify: Missing pages return 404/410, not 200
• Verify: Crawl waste from filters, calendars and parameters is controlled
• Verify: hreflang is reciprocal and self-referencing for localised pages
• Verify: Core Web Vitals meet targets at p75 on mobile field data