Recipe pages emit no JSON-LD

Recipe pages carry no JSON-LD: no rich result.

Current State

Measured over public/ of the current build, and over the 58 bundles under content/post/recipe/, every one of them type: recipe.

• No page carries an application/ld+json block at all, so a recipe is indexed as a plain page: no recipe rich result, and no ingredient list, step list, time or yield read by a crawler. The whole site sits at 0 of 100 in a structured-data audit.
• Everything a Recipe needs is already on the page: the title, the cuisine, the categories and tags, the portions, the prep-time and cook-time ISO 8601 durations, the author, and the ingredients.yaml and method.yaml sidecars that the two columns render.
• 12 of the 58 bundles carry a featured.webp of their own; the other 46 carry none, and the 15 section photographs of the cuisine pages are not a recipe’s image: images/featured-image.html stopped borrowing a section’s photograph for a page for that reason.
• The card is-structured-data-and-social-meta-tags covers structured data for the whole site, and asked for a Recipe block among the types it lists. This card is that one type, taken to the end, with a check.

Work

• Recipe partial - be-best/layouts/_partials/recipes/jsonld.html, the sibling of ingredients.html and method.html: it builds one Schema.org Recipe and writes it as <script type="application/ld+json">. The layout calls it, so the page’s structured data and its markup are built from the same values.
• What it states - name, url, @id as url#recipe, mainEntityOfPage, image, description, datePublished, dateModified, author, publisher, recipeYield, recipeCuisine, recipeCategory, keywords, inLanguage, prepTime, cookTime, totalTime, recipeIngredient and recipeInstructions. The three fields Google’s rich result requires are always there, and an optional key is merged in only where the content carries a value, so no key reads null or "" to a crawler.
• Nothing invented - a field with no value in the content is left out. image is the bundle’s own featured.webp, and the 46 pages whose bundle has none state no image; author falls back to the site’s author, which one recipe of the 58 needs; description falls back to the page’s summary, Hugo having no summary in .Description.
• Steps as HowToStep - Google asks for structured steps in place of an array of raw strings: each step is a HowToStep with position in the one sequence of the recipe, text, and name for its dish, under the rule recipes/method.html sets for the dish heading - a name where a recipe has more than one dish, and never the page’s own title twice.
• Times - prepTime and cookTime pass the front matter’s ISO 8601 durations through; totalTime is their sum, written from the minutes the two add to, the way recipes/recipe-times.html adds them for the table the page shows. PT1H15M stays PT1H15M, and PT0M adds nothing.
• Plain text - be-best/layouts/_partials/recipes/plain-text.html reduces a sidecar’s markdown to the words the page shows, so the link a step carries arrives as the text it shows and its markup does not. The substitutions of goldmark’s smartypants are written back as the characters the source carries, as the entity a build marks them with and as the literal character another build writes.
• safe.JS and safe.HTML - two marks the build needs, and neither is decoration. In a script element html/template writes a value it does not trust as a quoted JavaScript literal, so jsonify alone emits the JSON as a string: safe.JS is what makes it data. Writing a partial’s output escapes HTML, so a reduction that has just decoded every entity is re-encoded on the way out: safe.HTML is what keeps it reduced.
• Placement - the block is the first call in recipe/page.html, in main, beside the recipe it describes: Google reads JSON-LD from the body of a page, and the page renders as it did, its markup untouched.
• Definition - assets/recipe-jsonld.yaml defines the block in the JSON Schema style of the other definitions: the required keys, the const of @context, @type and the types inside the block, a pattern for an ISO 8601 duration and a language tag, minLength: 1 on every text field, and additionalProperties: false throughout, so a key the definition does not describe is a deviation.
• Check - pnpm run test:jsonld, scripts/validate-recipe-jsonld.mjs, in test:checks and so in the publish gate: it validates every block against the definition, then reads the block back against the bundle - ingredients and steps in file order, the times and their sum, the yield, the cuisine, the categories, the tags, the dates, the author, the image, and a description that stands in the page’s own prose. It scans the other 663 built pages for a Recipe block that no page shows, and reports any entity the reduction has no rule for.
• Method check, prose - validate-recipe-method.mjs asks that no step stands in the prose beside its lists. A block states every step as text, so it cut the page’s script elements out with the lists, as the prose a reader sees is what the check means.
• Definition of the contract - content/specification/technical/index.md gains Recipe Structured Data under Data Definitions and Recipe JSON-LD Validator under Tooling, and names the eight checks of test:checks; AGENTS.md names the partial, the definition, the check and the marks.

Acceptance Criteria

• Every recipe page carries exactly one application/ld+json block, of @type Recipe, validated against assets/recipe-jsonld.yaml: 58 pages, no deviation.
• The block states the sidecars: 698 ingredient lines and 442 steps over 77 dishes, in the order the files write them, each step a HowToStep at its position, none missing and none invented.
• The times are the page’s: prepTime and cookTime equal the front matter’s durations on all 58 pages, and totalTime is their sum.
• The block states no fact the page does not show: every description is prose of its own page, and every image is that bundle’s featured.webp, on the 12 pages that have one; the 46 pages with no photograph state no image.
• No page outside the recipe tree carries a Recipe block: the other 663 of the 721 built html files scanned, none found.
• The page is undisturbed: the same 426 pages with exactly one h1, the same 45 pages with images against the same 59 placeholder classes and 59 rules, and html-validate clean over the 721 html files, the blocks and the script elements included.
• pnpm run test passes, and pnpm run test:checks passes over the minified production build the publish gate runs.
• Google’s own warnings, listed: a page with no image warns Missing image, so 46 pages warn; the publisher is a Person, so Missing publisher logo warns on every recipe page; video, cost, calories and the rest of Google’s recommendations are fields the content does not carry, and the block does not state them. The Rich Results Test is a web tool and was not run; the check enforces the fields it reads.

Notes

• @id is url#recipe and mainEntityOfPage names the page, so the recipe is the entity of the page, not a copy of it, and a crawler that merges the two finds one recipe.
• dateModified equals datePublished on every page: the site builds with no enableGitInfo, so Hugo’s .Lastmod falls back to the page date. A build that turns git info on makes the field mean what it says, and needs no change here.
• inLanguage is read from the site’s locale: en-us and written as en-US, the case BCP 47 asks of a region subtag.
• Hugo’s jsonify escapes <, > and & as unicode escapes, so no ingredient or step text can close the script element, and it orders a map’s keys, so a rebuild writes the same bytes.
• html/template drops HTML comments from a layout at parse time: the comments recipe/page.html writes, Left Column among them, are in the template and never in the page. Debug output from a layout has to be markup, not a comment.
• The check compares text with markup stripped, entities decoded and whitespace dropped for the description: the minified build leaves no space at a paragraph break, where a summary spans one, and both builds then pass the same check.

Not in scope

• The rest of is-structured-data-and-social-meta-tags: WebSite, BlogPosting, BreadcrumbList and AboutPage, the og:type of an article, socialImage, and the robots, author, publisher, creator and keywords meta tags.
• A featured image for the 46 recipe bundles that carry none, which is what clears their Missing image warning.
• Fields the content has no value for: nutrition and calories, cost, video, a yield or a time for one component dish, and a per-dish image.
• A block for a cuisine page: a section lists recipes, and a Recipe names none of them.