IWAC SEO

Search-engine optimisation and XML sitemap for the Islam West Africa Collection (IWAC) Omeka S instance (islam.zmo.de).

The site ships almost no SEO metadata out of the box: pages have a <title> but no description, no Open Graph / Twitter cards, no canonical link, no structured data, no Google Search Console verification, and there is no sitemap.xml or robots.txt. This module fills all of that in — manually for static pages, automatically for every resource page — and adds a sitemap, a robots file, and an optional IndexNow ping.

It is self-contained, with no third-party runtime Composer dependencies or theme edits. Version 1.1 adds a durable IndexNow outbox table. See operations and migration for the server cron and optional Search Console workflow.

IWAC is bilingual. The same collection is published as two Omeka sites — afrique_ouest (French, the default the host root redirects to) and westafrica (English). The module resolves the current site at request time, so every signal (site title, locale, self-referential canonical, breadcrumbs) is emitted in the right language, and each page advertises its other-language counterpart via rel="alternate" hreflang. See Bilingual SEO.

---

What it does

Area	Detail
Meta tags	<title>, <meta name="description">, canonical link on every public page.
Open Graph / Twitter	og:title/description/image/type/url/site_name/locale + twitter:card/title/description/image/site so shared links render a rich preview.
schema.org JSON-LD	Per-resource structured data typed from the resource class (Person, Place, Organization, Event, NewsArticle, PublicationIssue, the scholarly reference types, VideoObject …), plus WebSite on the home page and BreadcrumbList on resource pages.
Citation metadata (Zotero)	Highwire Press citation_* + Dublin Core DC.* <meta> tags so the Zotero Connector, Google Scholar, Mendeley and other reference managers capture each item as a properly-typed reference (newspaper article, magazine issue, journal article, book, chapter, thesis, report, blog post …).
unAPI (Zotero RDF)	Primary-source items also advertise a /unapi endpoint serving Zotero RDF. Zotero prefers unAPI over the meta tags, so it imports a fuller record — the call number (Cote) from the iwac- identifier, single-field institutional creators, and Sujet + Couverture spatiale as tags.
Item-page citation tools	A "How to cite" resource page block — a formatted Chicago / APA / MLA reference (switchable, copy-to-clipboard) plus BibTeX / RIS / CSL-JSON downloads at /cite/{id}/{format} and the Zotero-RDF link for eligible kinds. Placed via the theme's Configure resource pages screen; the theme renders the UI (its common/citation partial) via the iwacCitation view helper, and this module owns the data. Replaces the BulkExport block for single-item exports.
og:image	The large thumbnail of the item's primary media (the page scan / cover); falls back to a site-wide default share image.
XML sitemap	/sitemap.xml index → /sitemap-pages.xml, /sitemap-item-sets.xml, /sitemap-items-{n}.xml (5,000 items per file by default). Public resources only, with <lastmod>, <changefreq>, <priority>. Cached.
robots.txt	/robots.txt disallowing /admin and pointing crawlers at the sitemap. The staging switch uses page-level noindex, which crawlers can fetch.
Google Search Console	Paste the verification snippet in the module config; the <meta name="google-site-verification"> tag is injected site-wide.
IndexNow ping	Optionally notifies Bing/Yandex when public content changes (durable, throttled batches and retries).

Static-page SEO is set by hand (a central admin table); resource-page SEO is derived automatically from each item's metadata, with the configurable defaults filling any gaps.

---

Why dispatch on resource class, not template

Both the schema.org typing and the citation typing key off the Omeka resource class (the RDF type), not the data-entry template. IWAC's templates are not 1:1 with classes:

• Template 8 ("Newspaper article") historically held both newspaper articles (class 36 bibo:Article) and Islamic-publication issues (class 60 bibo:Issue); the latter now also has its own template 21, but class is still the reliable split.
• The bibliographic references share templates across classes (e.g. template 10 covers bibo:Book, bibo:EditedBook and bibo:Thesis).

So the maps in config/module.config.php are keyed by class id. This matches how the IWAC → Hugging Face pipeline derives its subsets (also by class). See the iwac-data skill (omeka-structure.md) for the full class catalogue.

---

Installation

The module folder must be named IwacSeo inside Omeka's modules/ directory (the folder name has to match the namespace).

Download IwacSeo-<version>.zip from the latest release and unzip it into modules/ — it already unpacks to a correctly named folder and contains only the runtime. Do not use GitHub's "Source code" archives: they unpack to IWAC-SEO-<version>/, which the namespace will not resolve. Cloning works too, provided you name the target directory:

git clone https://github.com/fmadore/IWAC-SEO.git modules/IwacSeo

Then:

1. Admin → Modules → IWAC SEO → Install.
2. Configure (see below) — at minimum paste your Google Search Console snippet and pick a default share image.
3. Visit /sitemap.xml and /robots.txt to confirm they render.

No web-server changes are needed: nginx's try_files … /index.php already routes the .xml/.txt endpoints to Omeka because no such static files exist.

There are no third-party dependencies, so composer install is not required — Omeka autoloads the IwacSeo\ namespace from src/.

Requirements: Omeka S ^4.2.0, PHP >= 8.2.

---

Configuration

Modules → IWAC SEO → Configure. Every field is optional; sensible defaults are applied on install.

Setting	Purpose
Google Search Console verification	Paste the whole <meta name="google-site-verification" …> tag (or just the token). Injected on every public page. Then add the property at search.google.com/search-console and submit /sitemap.xml.
Bing Webmaster verification	Same, for the msvalidate.01 tag.
Default meta description	Used on pages without their own description (home, browse, search).
Default social share image	og:image fallback (≈1200×630). Used when a page or an item with no media is shared.
Twitter / X @handle	Emitted as twitter:site.
Discourage indexing (whole site)	Staging switch: every page becomes noindex,nofollow while robots.txt permits reading that directive. Off in production.
Noindex filtered browse	Keeps filter/sort URLs out of the index; clean pagination stays indexable and self-canonical. On by default.
Emit schema.org JSON-LD	Toggle structured data. On by default.
Emit citation meta tags	Toggle Zotero / Scholar tags. On by default.
Serve unAPI (Zotero RDF)	Toggle the /unapi endpoint + discovery tags for primary sources. On by default; needs the meta tags on for the fallback.
Serve /sitemap.xml & /robots.txt	Toggle the sitemap/robots endpoints. On by default.
Sitemap cache lifetime	Seconds before the cached sitemap is rebuilt. Default 86400 (24h).
Ping IndexNow when content changes	Off by default. See the caveat below.
IndexNow key	A hex key you choose (openssl rand -hex 16); served at /{key}.txt.

Per-page SEO

Admin → SEO → Static pages lists every site page with editable meta title, meta description, share image (picked with Omeka's asset selector) and indexing (default / index / noindex). Blank fields fall back to the page's own title and the site-wide defaults. Resource pages are not listed — their SEO is automatic.

Because IWAC publishes the collection as two Omeka sites (afrique_ouest/fr and westafrica/en), the screen opens on the default site and a Site selector at the top switches between them — overrides are stored per site, so the English pages (home, about, …) are edited on the westafrica site and the French ones on afrique_ouest.

Admin dashboard

Admin → SEO shows what is configured, the sitemap/robots URLs with public-resource counts, a bilingual (hreflang) coverage report, and a Regenerate button that clears the sitemap cache so it rebuilds on the next request. The coverage report separates the two ways a page can be wrong about its translations: broken alternate links — a page_pairs row naming a counterpart that is not a public page, so the alternate 404s — and pages with no pair, which simply emit none. Broken is listed first because it is the worse of the two. It also warns when the stored IndexNow key cannot match the /{key}.txt route (non-hex) and would fail verification.

---

How the metadata is produced

The head signals are written into Omeka's request-global head placeholder helpers (headTitle, headMeta, headLink, headScript), which the theme already echoes in <head> — so no theme template needs to change.

• Resource pages (view.show.after on the Item/ItemSet/Media controllers): title from displayTitle(), description from bibo:shortDescription (the AI summary on newspaper articles) / dcterms:abstract / dcterms:description (truncated to ~160 chars), canonical from the resource's site URL, og:image from the primary media, and JSON-LD from the resource class.
• Static pages (view.show.after on the Page controller): the editor's per-page overrides, else defaults; WebSite JSON-LD on the home page.
• Browse/search pages (view.browse.after): self-referential canonical and optional noindex on filtered variants; tracking parameters are removed and pagination remains indexable.
• Every page (view.layout): site-wide constants (og:site_name, og:locale, twitter:card, verification tags) and gap-fills for anything not already set. Resource values always win because the resource listeners run before the layout listener.

The two canonicals differ deliberately. On the browse routes this module owns, page 2 of a listing is a real page in a series, so its canonical is self-referential and noindex is what keeps it out of the index. On a route no phase-1 listener claimed — the IwacSearch /search app, any other module's controller — the layout gap-fill has no such knowledge, so it canonicalises to the query-less URL and marks any query-carrying variant noindex, follow. og:url mirrors whichever canonical was written, so a share never disagrees with it.

Turn off Omeka's own JSON-LD embed. Independently of this module, Omeka S core echoes AbstractResourceRepresentation::embeddedJsonLd() — the resource's entire API representation — into the body of every resource show and browse page. On IWAC that is 157 KB per item (it includes the OCR text) and 2.2–3.9 MB on an authority record, whose @reverse lists all ~13,000 items linking to it; Google truncates it and reports unparsable structured data. Disable it per site under Admin → Sites → … → Settings → General → "Disable JSON-LD embed". The data stays available at /api/items/{id}, and Zotero import here uses unAPI and the citation meta tags above, not the embed.

schema.org @type map

Driven by config/module.config.php → iwac_seo.structured_data.class_types (overridable via config/local.config.php), keyed by Omeka resource class id:

Resource class	@type
foaf:Person (94)	Person (+ givenName/familyName, affiliation)
dcterms:Location (9)	Place (+ geo from curation:coordinates)
foaf:Organization (96)	Organization (+ parentOrganization)
bibo:Event (54)	DefinedTerm ‡
fabio:AuthorityFile (244)	DefinedTerm (subjects / authority files)
bibo:Article (36)	NewsArticle (newspaper article)
bibo:Issue (60)	PublicationIssue (Islamic-publication issue)
bibo:AudioVisualDocument (38)	VideoObject (+ duration, uploadDate, thumbnailUrl, embedUrl/contentUrl)
bibo:Document (49)	DigitalDocument
bibo:AcademicArticle (35)	ScholarlyArticle
fabio:BookReview (178)	ScholarlyArticle (+ about = the reviewed book) †
bibo:Chapter (43)	Chapter
bibo:Book (40) / bibo:EditedBook (52)	Book
bibo:Thesis (88)	Thesis
bibo:Report (82)	Report
bibo:PersonalCommunication (77)	CreativeWork (+ isPartOf = the event's URL, when it is a record)
fabio:BlogPost (305)	BlogPosting

† Book reviews are not typed Review. Google's review snippet requires reviewRating.ratingValue, and an academic book review awards no score, so a Review node is reported invalid in perpetuity — an error bought for a feature that can never appear. As a scholarly article about a book, the relationship survives in a form that validates; the Zotero/Highwire side is unaffected and still declares the review citation kind.

‡ Events are not typed Event, for the same reason book reviews are not typed Review. Google's Event feature is for events "bookable to the general public" and asks for offers, performer, organizer, eventStatus and a PostalAddress. All 243 of IWAC's event records are Notice d'autorité — historical congresses and conferences, none of them attendable — so they are ineligible by that guideline however complete their metadata, and the type buys 97 errors and some 500 warnings for a rich result that can never appear. DefinedTerm is what the archive calls them and what class 244 already uses; the 115 Wikidata sameAs links that do the entity-resolution work are unaffected.

For the same reason a conference paper's isPartOf no longer builds an Event node. That node was out of range twice over: isPartOf ranges over CreativeWork or URL, and the event was ineligible anyway. Six of the nineteen papers name a linked event record and keep the cross-link as its URL; the ten holding only a literal title emit nothing, a name Google cannot dereference not being worth a range violation.

The Event branch itself is kept in StructuredData because class_types is overridable in config/local.config.php — an installation whose events are bookable wants that shape, and it handles the interval case: a dcterms:date may be an ISO 8601 interval (1979-11-04/1981-01-20, on 55 of the 243 records), which becomes startDate + endDate, with either side that is not a plain ISO date dropped rather than emitted, since a date a validator cannot read invalidates the node around it.

thumbnailUrl is taken only from a picture of the resource itself, never from the site's default share image — unlike image, which may fall back to it. A share card needs some picture; a claim that the site logo depicts the video is a different kind of statement. Two sources qualify, in this order: an asset assigned as the item's own thumbnail (the editor's explicit choice, and the way to give a video whose file yielded no still one that is actually of it), then the primary media's derivative. A media without derivatives is skipped, because Omeka answers for it with its generic file-type icon (video.png), which is a picture of nothing in particular.

A video with no description of its own — no dcterms:description, dcterms:abstract, bibo:abstract or bibo:shortDescription, which is 310 of the 1,790 — gets one composed from the record's own facts, in the page's language, since description is required of a VideoObject and Search Console reports every omission as an error:

Enregistrement vidéo (2 min 49 s) publié par RTB - Radiodiffusion Télévision du Burkina le 15 avril 2022, en français. Lieux : Burkina Faso. Collection Islam Afrique de l'Ouest.

That is a catalogue entry, not prose: who made it (bibo:authorList / dcterms:creator), who published it, when (to the archive's own precision — a year stays a year), how long it runs, in what language, where and about what, then the collection. Nothing in it is inferred, which is the line 1.0.4 drew against summarising the machine transcript. A record that carries any descriptive text keeps it untouched, and a record with no fact beyond its title gets no sentence rather than an empty one. VideoDescription holds the wording, in the same EN/FR string-table style as the citation formatter — the module's services run without a translator.

A video's uploadDate is emitted only from an explicit dcterms:issued timestamp with a known time zone. Catalogue dates retain their precision in datePublished; missing days, times and zones are never invented. Incomplete video metadata remains an editorial task and may prevent rich-result eligibility.

embedUrl and contentUrl say where the video can be played, and come from disjoint sources: 1,745 records name a source in fabio:hasURL — all but two a YouTube watch page, which becomes the /embed/ form — and the 44 digitised from DVD hold the file itself as an item media, whose original URL is the contentUrl. Only a media Omeka serves as video/* qualifies: contentUrl must point at content bytes, so a cover image is not a candidate and neither is a private media. A source URL that resolves to no player (the collection holds one SoundCloud track and one Wayback capture) yields nothing rather than a link Google cannot play.

Creative works also carry author/editor/contributor (from bibo:authorList, dcterms:creator, bibo:editorList, dcterms:contributor), datePublished, inLanguage, keywords, spatialCoverage, isPartOf (the journal/newspaper/book/event) and publisher when present. Every record carries sameAs from its dcterms:identifier URI values (Wikidata, GeoNames, VIAF …) — opaque internal ids like iwac-article-0000001 are skipped.

---

Citation metadata (Zotero, Google Scholar, …)

So that the Zotero Connector and similar tools capture a resource page as a reference, each item page also emits embedded bibliographic <meta> tags — the two vocabularies Zotero's Embedded Metadata translator reads:

• Highwire Press (citation_title, repeated citation_author / citation_editor, citation_publication_date, citation_journal_title / citation_inbook_title, citation_volume, citation_issue, citation_firstpage/citation_lastpage, citation_publisher, citation_dissertation_institution, citation_technical_report_institution, citation_doi, citation_language, citation_keywords, citation_abstract, citation_pdf_url, citation_public_url).
• Dublin Core (DC.title, DC.creator, DC.date, DC.publisher, DC.type, DC.language, DC.identifier, DC.subject, DC.description) — emitted for every resource, including authority pages, as a generic fallback.

IWAC's field conventions are baked into the per-kind tags (see src/Service/CitationMeta.php):

• the journal / newspaper / publication title lives in dcterms:publisher (often a linked item set), not dcterms:isPartOf;
• a book chapter's book title lives in dcterms:alternative;
• DOIs live in bibo:doi (a URI value); ISBN/ISSN are not recorded in IWAC, so those tags are not emitted;
• Zotero tags are built from both dcterms:subject (Sujet) and dcterms:spatial (Couverture spatiale). Both sets are emitted as repeated DC.subject — the channel Zotero's Embedded Metadata translator turns into tags via its RDF backend (it pre-empts citation_keywords) — and also folded into citation_keywords for Google Scholar and as a fallback.

The citation kind per class lives in iwac_seo.citation.class_kinds and is overridable. citation_pdf_url is emitted only for a public PDF media, so restricted bitstreams are never exposed. Toggle the whole feature with Emit citation meta tags in Configure.

Newspaper articles & magazine issues — correct Zotero typing

Newspaper articles (class 36) and Islamic-publication issues (class 60) are the bulk of the archive, but Highwire Press has no container tag for a newspaper or magazine — and any citation_* container tag forces Zotero's item type to journalArticle (in the Embedded Metadata translator, the Highwire type wins over DC.type). So those kinds instead:

• set DC.type to a valid Zotero item-type id (newspaperArticle / magazineArticle), which the translator's RDF backend honours because no Highwire container tag is present; and
• route the publication name through prism.publicationName → Zotero's publicationTitle.

The same DC.type technique types blog posts (blogPost), audiovisual (videoRecording) and scientific communications (presentation). The scholarly references (journal article, book, chapter, thesis, report, review) use the standard Highwire container tags, which type them precisely on their own.

unAPI — full-fidelity Zotero import for primary sources

Two things the archive needs simply cannot be expressed through the flat <meta> tags that Zotero's Embedded Metadata translator reads (verified against translators/RDF.js):

• a call number (French Zotero: Cote) — Zotero only fills callNumber from a typed RDF node (dcterms:LCC), never from a meta tag; a bare dc:identifier that is not an ISBN/ISSN/DOI is dropped; and
• a single-field institutional creator — a literal author is always run through Zotero's cleanAuthor() and split, so Association Islamique d'Al Mawadda Burkina Faso becomes … Burkina / Faso.

So the primary-source kinds (newspaper article, periodical issue, document, audiovisual, photograph) additionally expose unAPI: each page carries <link rel="unapi-server" href="/unapi"> and <abbr class="unapi-id">, and /unapi?id={url}&format=rdf_zotero serves the item as Zotero RDF. Zotero ranks the unAPI translator (priority 300) above Embedded Metadata (400), so the Connector imports the RDF instead of scraping the meta tags — which lets the module set every field exactly:

• z:itemType → the precise item type; dc:identifier → dcterms:URI → the item URL;
• Cote ← the iwac- identifier, as a dc:subject → dcterms:LCC → rdf:value node → callNumber;
• institutional creators ← a foaf:Person node with only foaf:surname → a single-field creator; persons stay literal dcterms:creator values that Zotero splits into first/last;
• tags ← dcterms:subject (Sujet) + dcterms:spatial (Couverture spatiale) as dc:subject;
• prism:publicationName, prism:number/prism:volume, bib:pages, dc:date, dc:language, dcterms:abstract, dc:rights, and the public PDF as an eprints:document_url attachment.

The Highwire / DC <meta> tags stay on every page — they still feed Google Scholar and are the fallback if unAPI is unreachable, and they remain the sole path for the bibliographic-reference kinds (which Highwire already types precisely). ZoteroRdf reuses the same iwac_seo.citation.class_kinds map as CitationMeta; toggle the endpoint with Serve unAPI in Configure (the iwac_seo_unapi setting).

---

Item-page citation tools — "How to cite" + downloads

The capture above is for reference-manager software. Human readers get a public "How to cite" panel, provided as a resource page block: this module registers it (resource_page_block_layouts), so it appears in the theme's Configure resource pages screen (Admin → Themes → your theme → Configure resource pages) and an admin controls its region and order. The block delegates to the theme (IWAC-theme) for the UI (its common/citation partial + styling + JS) while this module owns the citation data, so the resource-class → kind mapping is never duplicated. The block reads its data through the iwacCitation($item) view helper, which returns:

• a formatted Chicago (default), APA and MLA reference — switchable, one-click copy — produced by CitationFormatter. It is hand-rolled (no CSL-processor dependency; the module ships no bundled vendor/) and bilingual: connectives and month names follow the site language, so the French site reads Dans, sous la dir. de, 7 décembre 2018;
• downloads at /cite/{item-id}/{format} — BibTeX (.bib), RIS (.ris) and CSL-JSON (.json), served by CitationController as an attachment whose filename is the iwac- accession id; and
• the Zotero RDF link (the /unapi endpoint above) for the Connector-eligible kinds.

This is the single-item replacement for Daniel-KM/Omeka-s-module-BulkExport. All three formatters and serialisers read one normalized record from CitationData, which reuses the CitationMeta field conventions (container in dcterms:publisher, a chapter's book title in dcterms:alternative, NumericDataTypes dates, person-vs-institution creators). Authority records (person, place, organisation, event, subject) are not citable works — the helper returns null and the panel is hidden. The panel and downloads share the citation kill-switch with the meta tags (the iwac_seo_citation_meta setting).

---

Bilingual SEO (hreflang)

IWAC is the same collection under two Omeka sites — afrique_ouest (fr) and westafrica (en). The module ties them together the way Google expects:

• Canonicals are self-referential per language. The French page canonicals to its own …/afrique_ouest/… URL, the English page to …/westafrica/…. It never points one language at the other (that would drop a language from the index).
• rel="alternate" hreflang links every page to its counterpart — with a self link and an x-default (the French site, matching the host-root redirect) — so Google serves the right language and treats the pair as one set instead of duplicate content.

  • Resources (items, item sets, media) are shared across both sites under the same o:id, so the alternate is just the same path under the other site slug — fully automatic.

  • Static pages have different slugs per language (accueil/home, a-propos/about …); they are mapped by the iwac_seo.hreflang.page_pairs table in config/instance.config.php. A page with no entry simply gets no alternate (never a broken one). Update the table when pages are added or renamed.

    The table is generated, not authored. The Internationalisation module records each page's counterpart and exposes it on the REST API as o-module-internationalisation:related_page, so that module is the only place a pairing is written down; the committed table is a build product, which is what keeps a page render free of any lookup. composer hreflang:fix regenerates it, and composer hreflang:check audits it — reporting pairs the module records but the table omits, rows that contradict the module (which emit a 404 alternate — worse than emitting none), and rows naming pages that no longer exist. Rows are sorted and aligned deterministically, so regenerating an already-correct table rewrites nothing.

    Adding or renaming a page is what invalidates the table, and nothing in this repository changes when that happens — so the hreflang drift workflow regenerates it weekly and opens a pull request if anything moved (a pull request, not a push: the table drives what search engines are told about every page). The same workflow audits, without rewriting, any PR that touches the table. It needs the network, so it is deliberately outside composer check.

• og:locale + og:locale:alternate advertise both languages to social platforms.
• The item / item-set sitemaps carry <xhtml:link rel="alternate" hreflang> for each language, so both versions are discoverable from the single (default-site) <loc> entries.

Configure the language map, x_default and page pairs under iwac_seo.hreflang (override via config/local.config.php); set enabled => false to turn it all off.

Sitemap

• /sitemap.xml — sitemap index listing the children below.
• /sitemap-pages.xml — the home page + all public site pages (driven by the site navigation, so menu order and depth set the priority).
• /sitemap-item-sets.xml — public item-set browse pages.
• /sitemap-items-{n}.xml — public items, defaulting to 5,000 URLs per file.
• /sitemap-pages-{site-slug}.xml — other public language sites, including untranslated pages; pages marked noindex are excluded.

Each item entry also carries an <image:image> element (the primary media's large thumbnail — the page scan or cover) so Google Images can index the scans; disable via iwac_seo.sitemap.include_images in a local config override.

Resource ids + modified timestamps are read with one lean DBAL query per type (public resources scoped to the site), avoiding representation hydration for each item. Output is cached under files/iwac-seo-cache/ and served with Cache-Control / Last-Modified headers; the cache is invalidated when an item or page changes, and any cache failure falls back to live generation.

---

IndexNow ping

Public content changes enter a durable outbox. Background jobs send leased batches of up to 200 URLs, grouped by origin, and acknowledge only accepted submissions. Failed work retries with backoff; leases expire after worker failure. Schedule the independent five-minute drain so pending work does not depend on another edit. See operations for migration, cron, retries and the optional GitHub Search Console monitor.

Google is not pinged: its sitemap-ping endpoint was retired in 2023. Google discovers content via the robots.txt Sitemap: line and Search Console.

---

Project layout

IwacSeo/
├── Module.php                       # listeners, ACL, install/upgrade/uninstall, config form
├── composer.json                    # type omeka-module; PSR-4 IwacSeo\ -> src/ (no runtime deps)
├── config/
│   ├── module.ini
│   ├── module.config.php            # wiring: routes, services, navigation
│   └── instance.config.php          # IWAC data: class maps + page translations
├── src/
│   ├── Controller/
│   │   ├── SitemapController.php     # /sitemap*.xml, /robots.txt, /{key}.txt
│   │   ├── UnapiController.php       # /unapi -> Zotero RDF
│   │   ├── CitationController.php    # /cite/{id}/{format} -> BibTeX/RIS/CSL-JSON
│   │   ├── Admin/SeoController.php   # dashboard + static-page table
│   │   └── Concern/SendsResponses.php  # shared XML/text/file response plumbing
│   ├── Form/{ConfigForm,PageSeoForm}.php
│   ├── Job/PingSearchEngines.php
│   ├── View/Helper/Citation.php      # iwacCitation() -> "How to cite" view-model
│   ├── Site/ResourcePageBlockLayout/Citation.php  # "How to cite" resource page block
│   └── Service/
│       ├── HeadMetadata.php          # decides every <head> SEO signal
│       ├── HeadWriter.php            # writes them; tracks what the request has set
│       ├── StructuredData.php        # schema.org JSON-LD (by resource class)
│       ├── CitationKind.php          # the kind vocabulary + export type tables
│       ├── CitationKindMap.php       # resource class id -> kind (one shared map)
│       ├── CitationMeta.php          # Highwire + Dublin Core citation tags
│       ├── ZoteroRdf.php             # Zotero RDF (served via unAPI)
│       ├── CitationData.php          # builds a CitationRecord from an item
│       ├── Citation/                 # CitationRecord, Creator, IssuedDate (the record)
│       ├── CitationFormatter.php     # Chicago / APA / MLA text (hand-rolled, bilingual)
│       ├── CitationExport.php        # BibTeX / RIS / CSL-JSON serialisers
│       ├── Concern/ResourceValueReader.php  # shared value-readers (trait)
│       ├── SitemapGenerator.php      # which URLs go in which sitemap
│       ├── Sitemap/                  # SitemapRepository, UrlsetWriter, XmlCache, SitemapDocument
│       ├── PageSeoStore.php          # per-page overrides (site setting)
│       ├── PingQueue.php             # IndexNow queue: dedupe, flood cap, throttle
│       ├── Pinger.php                # IndexNow submit
│       ├── SettingsGate.php          # typed reads over the iwac_seo_* settings
│       ├── ResourceUrl.php, ViewLocale.php, Text.php  # small shared helpers
│       └── *Factory.php
├── view/iwac-seo/admin/seo/{dashboard,pages}.phtml
├── asset/css/admin.css
├── language/                        # template.pot + fr.po/fr.mo
└── tests/                           # unit + real-Omeka integration suites

---

Verifying a deployment

1. curl -s https://islam.zmo.de/sitemap.xml | head → a <sitemapindex>; the child sitemaps list <url> entries with <lastmod>.
2. curl -s https://islam.zmo.de/robots.txt → Disallow: /admin/ and a Sitemap: line.
3. View-source an item page (/s/afrique_ouest/item/2231): confirm <title>, description, og:*, twitter:*, <link rel="canonical"> and an application/ld+json block. Validate the JSON-LD with the Rich Results Test.
4. Open a reference page (e.g. a journal article) with the Zotero Connector and confirm it saves as the right item type with authors, date, container and DOI; open a newspaper article and confirm it saves as a newspaper article with the newspaper as the publication.
5. Home page: og:site_name and a WebSite JSON-LD block; the GSC tag once a token is set.
6. In Search Console: add the property, verify via the meta tag, submit /sitemap.xml.

---

Roadmap

• COinS (<span class="Z3988">) as a complementary reference-embedding signal, and to expose references on list pages.
• Per-URL hreflang in the pages sitemap too (static-page alternates are already emitted on-page; only the item / item-set sitemaps carry <xhtml:link> so far).
• Optional nginx-level caching of /sitemap*.xml (the module already emits Cache-Control/Last-Modified; the image-sitemap entries shipped in 0.6.0).

---

Uninstalling

Removes every iwac_seo_* global setting, the per-site iwac_seo_pages overrides and the sitemap cache directory (files/iwac-seo-cache/).

---

Development

No production dependencies; composer install pulls the PHPUnit, PHPStan and PHP_CodeSniffer development tools. Run every local quality gate with:

composer install
composer check

The suite covers the pure logic that regresses most silently — the Chicago/APA/MLA formatter, the BibTeX/RIS/CSL-JSON serialisers, hreflang resolution and the text utilities, plus sitemap policy and IWAC's instance-configuration contracts. GitHub Actions (.github/workflows/ci.yml) runs syntax checks and PHPUnit on PHP 8.2–8.5, then checks Composer metadata, translations, PSR-12 and PHPStan at the declared PHP 8.2 floor. A separate PHP 8.5 job installs Omeka S 4.2.1 and exercises CitationMeta, ZoteroRdf, HeadMetadata, the service/controller managers and public routes against the real Omeka and Laminas classes. It intentionally does not add framework dependencies to this module or expand the narrow unit-test shims.

To run that boundary locally, install Omeka separately and point OMEKA_PATH at its root:

OMEKA_PATH=/path/to/omeka-s composer test:integration

Omeka's vendor/autoload.php is loaded before the module's development vendor, matching the production ownership of Laminas and PSR classes. ROADMAP.md documents the refactoring plan behind 0.6.0 and what remains deliberately out of scope.

---

Licence

GPL-3.0-or-later. © Frédérick Madore.

---

Citing

Use the Cite this repository button on GitHub, or CITATION.cff directly — it carries the ORCID and the released version, and GitHub renders it as APA or BibTeX. Keep its version and date-released in step with config/module.ini when releasing; the release workflow fails the build if they disagree.