Publication checks

Available with Core 0.2.22 and its updated organization build engine. This guide applies to customer-owned publishing on Cloud and self-hosted installations that use the frozen publication runner. A standalone Astro build or the legacy managed hosting runner does not automatically run this gate.

What happens when you publish

After the static build, Typeroll checks the complete output: new and reused HTML, JSON-LD, resolved routes, redirects, sitemaps, robots directives and rendered shared partials. This happens before activating the deployment. A technical failure keeps the current live site unchanged. Publishing readiness checks your configuration and exportability; ready=true cannot predict the result of a future HTML build.

The publication status includes seo_report in the portal, deployment API and get_deploy_status. Each issue includes a URL, generated file and line, a field or block ID when available, and suggested remediation. Shared partial issues carry their block ID on every affected page. Review the counts as well as the examples: reports retain up to 100 examples per severity and have a bounded size.

Technical failures include broken internal page targets, redirect loops, malformed JSON-LD, unsupported schema mappings, incorrect generated breadcrumb routes, conflicting canonicals, invalid sitemap routes and unnamed linked images. Explicit canonical overrides are checked against your configuration, including cross-domain originals. An intentionally noindex page may stay public and linked; it simply stays out of the indexable sitemap. Fragment filters and client-created anchors are not treated as new pages.

Review warnings without rewriting facts

Warnings cover repeated title words, duplicate or missing metadata, heading jumps, generic link text, internal redirects, unexpected nofollow and repeated destinations within one list. An image link and title link in the same card are one context; useful links repeated in separate navigation or content contexts are allowed. These warnings do not block activation and do not rewrite your content.

In Site settings → Editorial publication checks, record forbidden internal markers, phrases that require review and notes for editors. Agents can use update_site_settings with the same seo_review object:

{
  "seo_review": {
    "forbidden_markers": ["wave-1", "internal research note"],
    "claims": [
      {
        "phrase": "compare earnings",
        "guidance": "Review this promise: this directory does not provide earnings comparisons."
      }
    ],
    "notes": ["Prefer fewer supported facts over unsupported claims."]
  }
}

These examples are project choices, not global forbidden words. Phrase matching is case-insensitive and deterministic; it is not a factual or semantic audit. Notes guide human and agent review. Every publication checks all output again, including legacy and unchanged pages, so changed constraints apply site-wide. There are no hard title/description length limits, word counts or required exact-match link phrases in this validator.

Indexing, breadcrumbs and cards

noindex and nofollow are independent page fields. A public page with noindex: true normally emits noindex,follow; add nofollow: true only when that is intentional. Site equivalents are sitewide_noindex and sitewide_nofollow. Private previews and blocked versions keep strict indexing and following restrictions. Do not remove intentional noindex just to pass a publication check.

Breadcrumb JSON-LD uses resolved routes, including content-type prefixes, explicit page paths and the trailing-slash policy. The last item follows the configured canonical. The homepage has no BreadcrumbList. Do not add a second manual BreadcrumbList to work around generated output.

For post cards, use meaningful alt text or image_alt_field. If a separately linked image has empty alt, Typeroll names the link from the card title, with “View page” as the last fallback when no title is supplied. Decorative images without links may keep empty alt. No extra link wrapper is added.

Schema mapping migration

Content-type schema_field_map supports direct properties, such as {"excerpt":"description"}. Dotted paths such as address.addressLocality and reserved prototype keys are rejected with a field-specific error on save. They are not converted into nested objects. Old unsupported mappings fail publication until corrected; underlying CMS fields are preserved.

Review and remove or replace unsupported mappings before publishing. Do not map a city string to address to suppress the error: it would misrepresent the data. For a typed nested PostalAddress, use deliberate page JSON-LD containing only known public address data, or the supported site-level Organization address settings. Do not manufacture empty addresses. The gate checks structural correctness, not every Schema.org vocabulary rule or eligibility for rich results.

Build engine and verification identity

After upgrading Core, open Publishing → Builds → Update build engine once for each organization/provider you use. The qualification build updates the trusted runner; it does not publish customer sites. Existing engines without the validation capability cannot admit new publications.

Domain-only preparation intentionally retains the last published source. If that snapshot predates publication checks, publish the main version with the current Core first, then prepare the domain change. Typeroll reports publication_renderer_update_required; it does not silently change the frozen renderer or publish unsaved/current content during domain preparation.

Every report is bound to a publication ID, frozen source hash, configuration hash, output file-tree hash and, on successful completion, transported artifact hash. The trusted runner checks the output digest before activation. Renderer source and validator versions are output dependencies; this release invalidates old HTML render caches. A previous successful SEO report is never reused.

Checks are offline and use the actual artifact. They do not prove external-link availability, DNS readiness, runtime app output, every dynamic anchor or access by a verified search-engine crawler. Robots inspection understands documented product-token fallbacks, including Google-InspectionTool → Googlebot; it does not spoof a bot. These checks establish technical correctness, not rankings or traffic improvements.