aia-website/www/reports/phase-6a-library-architecture.md

Phase 6A Library Architecture Proposal

Decision date: 2026-10-09 UTC Starting checkpoint: 34d5394b3ce765028a27b9cddd9952de3f09bfe7 Status: proposal only; implementation is not authorized

A. Executive Decision Summary

Recommended architecture

1. Search: generate one compact JSON document index per language from the existing publication manifest and blog collection. Load it on demand and use a small dependency-free TypeScript client for normalization, ranking, result rendering, URL state, and accessibility. Search title, description, headings, category, author display data, and future tags. Do not index full article bodies in the first release.
2. Library pagination: render ten articles per static page, matching the verified production page size. Keep each locale's existing Library root as page 1, generate only nonempty /page/{n} pages for n >= 2, and permanently redirect /page/1 aliases to the corresponding root.
3. Author archives: derive localized author pages and their pagination from author records plus the same eligible-post catalog. Generate no manually maintained post lists. Author roots use explicit author translationKey relationships; numbered author pages do not become translations merely because their numbers match.
4. Routing and policy: add archive/search descriptors to the shared route and publication model rather than creating a second eligibility system. The existing [...slug].astro manifest-driven route architecture should dispatch content and generated archive kinds from one validated static-path inventory.
5. Progressive enhancement: server-rendered paginated Library and author pages remain complete navigation without JavaScript. Search is an explicit enhancement; without JavaScript the search control is not presented as functional and visitors retain normal archive navigation.

This is preferable at the present scale because the three measured metadata indexes total 42,090 bytes uncompressed, 12,693 bytes gzip, or 10,709 bytes Brotli for 65 records. English alone is 26,391/8,036/6,652 bytes. A custom metadata index keeps queries local, adds no package or service, excludes drafts at build time, and can implement Spanish/Arabic normalization explicitly. A full-body prototype would be 474,623 bytes uncompressed and 135,796 bytes gzip; that additional payload is not justified by a verified visitor requirement.

Important tradeoffs

• Metadata search will not find a term appearing only in article body copy.
• Dependency-free ranking will be deterministic and understandable but less sophisticated than fuzzy/stemmed third-party search.
• Production currently has 55 English articles and six ten-item pages; Astro has 46 English records and can create only five nonempty pages. The nine missing Phase 6C articles must be reconciled before English page 6 and author page 6 can be accepted. The architecture must not create an empty page or redirect a distinct archive page merely to satisfy a route count.
• The current localized author records are structurally usable but their names and English-only short descriptions do not reproduce the current Spanish and Arabic archive headings/biographies. Author content needs source-supported reconciliation and human language review before those archives publish.

Approval gates

Approval is required for the metadata-only search scope, the /search compatibility behavior, inclusion of accessible noindex syndicated records in listings/search, pagination/archive indexing and sitemap treatment, author record changes, and sequencing Phase 6C before final English pagination.

B. Current Architecture Inventory

Rendering and routing

• src/pages/[...slug].astro asks getPublicationManifest() for every eligible page/post route and dispatches the three slug: library records to separate English, Spanish, and Arabic components. There is no generated archive kind.
• LibraryEnglish.astro, LibrarySpanish.astro, and LibraryArabic.astro each independently filter the manifest by language, sort by date descending, render every eligible post, and repeat a five-item “popular” slice. English alone renders an inert <input type="search">; no script, form, filtering, state, or announcements exist.
• BlogIndexLayout.astro supplies shared hero/intro/sidebar layout and delegates list contents. BlogCard.astro creates localized article URLs and dates, but hardcodes Rula Diab's English byline/avatar and does not link to an author archive.
• BlogPostLayout.astro likewise hardcodes the visible English byline. Its BlogPosting JSON-LD independently resolves an exact locale/slug author and already fails for missing or ambiguous records.
• The three Library Markdown bodies contain old extracted listing snapshots. [...slug].astro bypasses these bodies, so they do not render, but they are a competing stale representation. Phase 6B should replace them with maintenance notes only after the generated Library implementation is accepted.

Publication and translation policy

• publication.ts combines page and blog collections and delegates route, canonical, eligibility, sitemap, collision, and translation validation to publication-policy.ts.
• Drafts are not generated. Published noindex and external-canonical records are generated but excluded from the sitemap and hreflang graph. Staging globally emits noindex,nofollow, an empty sitemap, and no hreflang.
• Translation equivalence is explicit and collection-scoped through translationKey; it is never guessed from slugs.
• Authors are intentionally outside the current publication manifest. Author records declare locale and translation keys, but no author routes exist.

Current content inventory

The historical 65-post count remains accurate, but it was independently recalculated from the current files and frontmatter:

Language	Records	Generated/non-draft	Noindex + external canonical	Sitemap-eligible	Raw Markdown/MDX body bytes
English	46	46	2	44	276,099
Spanish	12	12	0	12	105,272
Arabic	7	7	0	7	41,832
Total	65	65	2	63	423,203

There are no drafts. The two exceptions are community-highlight-meet-rula-diab and new-aia-scottsdale-office; both are accessible local English routes with verified noindex and external canonicals. Current Library components use eligible, not sitemapEligible, so all 65 appear in lists.

Every record has title, description, date, author, category, featured image, canonical, and translation key. All use category Library, all have an empty tags array, none has updatedDate, and none supplies image alt frontmatter. The content model can support tags and modification dates later without changing the archive algorithm. Search should not pretend empty tags add useful ranking.

Equal-date sets exist in both English and Spanish. Current ordering is date descending with an implicit stable route order inherited from the manifest. The new catalog should make this explicit: date descending, then normalized route/slug ascending.

Author inventory and relationships

Language	Author records	Author slug	Referencing posts
English	1	rula-diab	46
Spanish	1	rula-diab	12
Arabic	1	rula-diab	7

The three records share translationKey: rula-diab, name Rula Diab, one avatar, and the same short English description. The schema has name, slug, optional description, optional avatar, lang, and translationKey. It is sufficient to identify and relate archives, but not to reproduce production's localized credentialed headings and biographies without content reconciliation. An optional locale-specific displayName is justified for archive presentation while preserving name as the person's name in BlogPosting schema. No multiple- author field should be added until a real multi-author record requires it.

C. Production Behavior and URL Evidence

Method and freshness

robots.txt was checked first. It allows the requested public paths. Twenty-nine bounded sequential GETs were made from 2026-10-09 18:01:51 through 18:02:17 UTC with no-cache request headers, no redirect following, no authentication, and no form submission. Requests covered representative roots, first-page aliases, last pages, first invalid boundaries, author equivalents, and search. One headless Chromium GET then allowed the existing search JavaScript to render one result page; it did not submit a form or write to production.

HubSpot responses variously reported HIT, MISS, or REVALIDATED and mostly identified October 3–4 prerenders. Cache-Control: no-store, no-cache, must-revalidate did not establish origin freshness. Findings are current public behavior from consistent responses, with potentially cached content. The older 2026-10-07 route report remains historical evidence for URLs not repeated.

Observed behavior

Family	Fresh result	Visible results	Navigation / metadata
/library	200, lang=en-us	10	Pages 1–6; page 6 has 5; page 7 is 404
/es/library	200, lang=es	10	Pages 1–2; page 2 has 2; page 3 is 404
/ar/library	200, lang=ar dir=rtl	7	One page; page 2 is 404
English author root	200	10	Pages 1–6; page 6 has 5; page 7 is 404
Spanish author root	200	10	Pages 1–2; page 2 has 2; page 3 is 404
Arabic author root	200, RTL	7	One page; page 2 is 404

All six /page/1 candidates return 200 and show the same ordered article links, title, and counts as their root rather than redirecting. English trailing slashes on /library/ and /library/author/rula-diab/ return 301 to their no-slash forms. The inspected archive/listing pages expose no canonical or robots meta; their pagination/author URLs are absent from the production sitemap.

The author roots display substantial localized bios and credentialed headings: English Rula Diab, Clinical Director, M.Ed, BCBA, LBA, a Spanish localized equivalent, and an Arabic localized equivalent. These are not represented by the current local author records and must not be invented from this report.

Production search

Library search forms use GET /search with term plus repeated type=BLOG_POST and type=LISTING_PAGE. The search page's own form uses q and also includes SITE_PAGE. Raw /search and /search?term=autism HTML contains no results; the existing HubSpot JavaScript loads them after page load.

A browser-level GET of /search?term=autism&type=BLOG_POST&type=LISTING_PAGE displayed 66 mixed- language results, ten per result page, and result offsets 0, 10, … 60. The first page included English and Spanish Library roots/posts, confirming that production search is global rather than scoped to the Library page's language. The search page has no useful document title (raw parsing resolves an SVG title), canonical, or robots directive. With JavaScript unavailable, production shows no results.

D. URL Disposition Matrix

Current below means the October 9 check; historical means the accepted October 7 route evidence. Proposed archive pages are self-canonical, indexable, and sitemap-eligible under the shared policy. /search is noindex and excluded.

Production URL or pattern	Purpose / language	Evidence and existing behavior	Proposed Astro behavior and preservation	Redirect / canonical / indexing	Phase 6B validation
/library	Library page 1, EN	Current 200, 10 results, pages 1–6	Preserve static page 1	Self-canonical; index; sitemap; existing EN/ES/AR root hreflang	Generated HTML, canonical, first 10 sorted posts
/es/library	Library page 1, ES	Current 200, 10 results, pages 1–2	Preserve static page 1	Self-canonical; index; sitemap; existing root hreflang	Locale cards/date/labels and two-page navigation
/ar/library	Library page 1, AR	Current 200, 7 results, RTL	Preserve static one-page archive	Self-canonical; index; sitemap; existing root hreflang	lang=ar dir=rtl; no nonexistent page 2
/library/page/1	First-page alias, EN	Current 200 duplicate	Preserve incoming URL through 301 to /library	Redirect only; canonical destination /library; no HTML/sitemap	Redirect map and no duplicate HTML
/es/library/page/1	First-page alias, ES	Current 200 duplicate	301 to /es/library	Redirect only	Redirect test
/ar/library/page/1	First-page alias, AR	Current 200 duplicate	301 to /ar/library	Redirect only	Redirect test
/library/page/2–/5	Library pages, EN	Current 200; /2 historical/current, /3–/5 historical 200	Generate nonempty ten-item pages from current corpus	Self-canonical; index; sitemap; no hreflang inferred	Route/card boundaries, prev/next, direct access
/library/page/6	Library last page, EN	Current 200 with 5 production posts	Generate only after Phase 6C supplies the missing records; do not create empty/redirected substitute	Self-canonical; index; sitemap once real	Phase 6C prerequisite; assert 5 real cards
/library/page/7+	Invalid future boundary, EN	Current /7 404	Generate only if eligible count later requires it; otherwise 404	No canonical/sitemap while absent	Beyond-last route absent
/es/library/page/2	Library last page, ES	Current 200 with 2	Generate static page	Self-canonical; index; sitemap; no page-number hreflang	Exactly 2 cards today
/es/library/page/3+	Invalid boundary, ES	Current /3 404	No route until count requires it	404 while absent	Beyond-last route absent
/ar/library/page/2+	Invalid boundary, AR	Current /2 404	No route until count requires it	404 while absent	Beyond-last route absent
/library/author/rula-diab	Author root, EN	Current 200, localized bio, pages 1–6	Generate from EN author record and posts after author content reconciliation	Self-canonical; index; sitemap; explicit author-root hreflang only	Author resolution, bio, links, first 10 posts
/es/library/author/rula-diab	Author root, ES	Current 200, Spanish bio, pages 1–2	Generate from ES author record; no invented copy	Self-canonical; index; sitemap; author-root hreflang	Human-reviewed source content and route test
/ar/library/author/rula-diab	Author root, AR	Current 200, Arabic bio, RTL	Generate from AR author record; no invented copy	Self-canonical; index; sitemap; author-root hreflang	RTL and human-reviewed source content
/library/author/rula-diab/page/1	Author first alias, EN	Current 200 duplicate	301 to author root	Redirect only	Redirect test
/es/library/author/rula-diab/page/1	Author first alias, ES	Current 200 duplicate	301 to ES author root	Redirect only	Redirect test
/ar/library/author/rula-diab/page/1	Author first alias, AR	Current 200 duplicate	301 to AR author root	Redirect only	Redirect test
/library/author/rula-diab/page/2–/5	Author pages, EN	Current/historical 200	Generate from current matching posts	Self-canonical; index; sitemap; no numbered hreflang	Same deterministic chunks as Library today
/library/author/rula-diab/page/6	Author last page, EN	Current 200 with 5	Generate after Phase 6C; no empty substitute	Self-canonical; index; sitemap once real	Phase 6C prerequisite
/library/author/rula-diab/page/7+	Invalid boundary, EN	Current /7 404	404 until count requires route	Absent from sitemap	Beyond-last route absent
/es/library/author/rula-diab/page/2	Author last page, ES	Current 200 with 2	Generate static page	Self-canonical; index; sitemap	Exactly 2 matching posts
/es/library/author/rula-diab/page/3+	Invalid boundary, ES	Current /3 404	No route until required	404	Boundary test
/ar/library/author/rula-diab/page/2+	Invalid boundary, AR	Current /2 404	No route until required	404	Boundary test
/search	Search shell, global	Current 200; client-rendered; malformed/absent SEO metadata	Preserve one static compatibility route using generated locale indexes and locale filter	Self-canonical path; noindex,follow; no sitemap/hreflang	Direct route, empty state, JS/no-JS behavior
/search?term={query} and repeated type	Legacy search state	Current form behavior; mixed-language results	Read term; safely ignore legacy type; preserve URL and client-render compatible results	Canonical /search; noindex; query escaped	Direct-load and XSS/Unicode tests
/search?q={query}	Search-page form state	Current search page uses q	Accept as alias input; normalize to one in-page state without network forwarding	Canonical /search; noindex	q and term equivalence test
/search?...&offset={10n}	Search pagination state	Browser verified offsets 0–60	Honor nonnegative multiples of 10 on compatibility route; invalid values normalize to 0	Canonical /search; noindex	Direct page, last/invalid offset tests
/{locale?}/library/ and author trailing slash variants	Normalization	English examples current 301; other variants not individually requested	Host-level 301 to no-slash normalized route	Redirect only	Redirect-map and deployment acceptance test

E. Search Alternatives and Comparison

Approach	Quality / languages	Payload and build	Accessibility / privacy	Maintenance	Decision
Generated metadata JSON + custom TypeScript	Explicit field weighting; deterministic Unicode normalization; Spanish accent and Arabic mark handling tested locally; no stemming/fuzzy by default	Current all-language index 42,090 raw / 12,693 gzip; one small client module; ordinary Astro static endpoints	Full control of semantic form/status/focus; queries stay in browser except legacy /search URLs; no third party	Small project-owned helper and tests; transparent failure modes	Recommended
MiniSearch-style client index	Built-in field boost, prefix, fuzzy matching, suggestions	Adds dependency and serialized index/client code; exact project payload not measured because no package was installed	Local/private and static; UI still project-owned	Better relevance features, but tokenizer/fuzzy behavior needs locale tuning and dependency upkeep	Defer until measured search-quality failures justify it
Pagefind-style post-build search	Indexes rendered pages after Astro, separates indexes by document language, supports static hosting	Adds post-build executable/runtime bundles and tends toward full-page text; changes build/deployment pipeline	Local/private; UI can be accessible but requires integration; no server	Strong general-site search, but broader than this Library and duplicates manifest eligibility/filter work	Reject for current 65-record Library; reconsider for verified site-wide full-text search
External hosted search/API	Potentially strongest ranking/analytics	Network/service dependency and operational credentials	Visitor terms leave the site; logging, consent, retention, failure, and cost concerns	Highest operational burden	Reject under static/data-minimization requirements

The proposed index stores route, locale, title, description, headings, category, future nonempty tags, author display name/slug, and publication date. It excludes article body HTML, unpublished content, likes/views, and sensitive or runtime data. Featured images need not be duplicated; result cards can use the existing article metadata or omit thumbnails in the compact search view.

Normalization should use Unicode NFKC, locale-aware lowercasing, collapsed whitespace, and Unicode punctuation separators. A secondary comparison key may fold Latin combining marks so evaluacion finds evaluación, and remove Arabic tashkeel/tatweel so marked and unmarked forms match, while preserving original display text. Do not apply English stemming to Spanish or Arabic. Rank exact title phrase, title tokens, heading matches, description, category/tags, and author in that order; break equal scores by date descending then route ascending. Prefix matching should require a minimum token length. Empty queries render the normal archive; no-result queries announce zero and offer clear/reset.

F. Pagination and Author Architecture

Shared catalog

Add a pure library-catalog module that accepts the existing publication manifest and author records and returns:

• eligible posts grouped by locale;
• explicit deterministic ordering;
• ten-item page slices and nonempty page descriptors;
• exact locale/slug author resolution with duplicate/missing validation;
• author archive descriptors derived from post relationships;
• compact locale search records.

It must not read extracted Library Markdown listing bodies or maintain manual post arrays. Current accessible noindex/external-canonical posts remain listable because that is established behavior, while drafts remain absent. An author archive should require a valid locale author record and at least one eligible post, including at least one sitemap-eligible post before the archive itself can be indexable.

Static route generation

Extend the manifest-driven static-path inventory with validated generated route descriptors (library-page, author-index, author-page, search) and run their canonical, collision, robots, and sitemap state through reusable Phase 3A policy functions. Do not hand-author content records for numbered pages. The catchall renderer can dispatch descriptor kinds to shared Library/archive components while preserving content routes and collision detection.

Page 1 is always the unnumbered root. Next/previous and numbered links are real anchors with localized accessible labels, aria-current="page" on the current page, and no links to nonexistent pages. Invalid numbers, zero, negative values, nonintegers, unknown author slugs, authors with no eligible posts, and pages past the end produce no static route and therefore 404.

Search temporarily replaces the visible page list with all-language-appropriate matching results and hides the archive pager; clear/reset restores the original server-rendered page. Adding, removing, drafting, or changing an article updates search and page chunks in one build without manual index edits.

Author presentation

Resolve the author once by exact locale and slug. Reuse the same resolution in BlogCard, BlogPostLayout, BlogPosting schema, and author archives. Add an optional locale-specific displayName only to preserve the published credentialed archive heading; retain name for the Person entity. The existing description/avatar can drive the bio after source reconciliation. Build must fail on missing or duplicate locale/slug records. Do not treat authors as team members or infer employment.

Route-count model

With the current 46/12/7 corpus and ten items per page:

Addition	HTML routes	Sitemap URLs	Hreflang links
/search	+1	0	0
Library pages 2+ (4 EN, 1 ES)	+5	+5	0
Three author roots	+3	+3	+12 if explicit author equivalence is approved
Author pages 2+ (4 EN, 1 ES)	+5	+5	0
Projected current total	96 + 14 = 110	92 + 13 = 105	174 + 12 = 186

This current-corpus model intentionally lacks the two distinct English page-6 routes. If all nine Phase 6C English posts are ordinary eligible/indexable records, those posts add nine HTML/sitemap URLs and produce one more Library and one more author page. The combined projection becomes 121 HTML routes and up to 116 sitemap URLs. Exact counts must be recalculated from final policy; new translation relationships could separately affect hreflang.

G. Multilingual and SEO Design

• Each JSON index contains one locale only. Embedded Library search never mixes languages. The /search compatibility page may search all locales but must expose an explicit locale filter and label result language.
• Components receive localized labels/messages; none are generated by machine translation. Missing copy blocks implementation for that locale rather than falling back silently to English.
• Arabic uses the existing lang=ar dir=rtl layout. Pagination order, arrows, search icon placement, result counts, focus order, and mixed Latin/Arabic text require RTL browser testing. Use Intl.DateTimeFormat/Intl.NumberFormat with the page locale; keep machine-readable dates stable.
• Existing Library roots retain their explicit three-language translation set. Numbered pages do not receive hreflang because page N contains different records across locales.
• Author roots may participate only through the author collection's explicit, validated translationKey, with reciprocal published archives and English as x-default. Numbered author pages receive no hreflang.
• /search has no translation graph and is noindex. JSON endpoints and redirect sources never enter the sitemap.
• Generated archive pages self-canonicalize to normalized local URLs and use the staging/production robots switch. Page 2+ and author archives are recommended as indexable and sitemap-eligible to remain consistent with Phase 3A's “all eligible self-canonical local routes” rule. This intentionally improves on production's missing canonicals and sitemap omission and requires approval.
• LanguageSwitcher uses only the validated graph. It must not construct page-N, author, or search equivalents from route shape.

H. Accessibility and Progressive Enhancement

• Render complete cards and archive pagination on the server.
• Hide/reveal the enhanced search form only after its script initializes; include a <noscript> explanation and links to the three Library roots. Do not show a control that appears functional when JavaScript is unavailable.
• Use an explicit label, native search input, submit button, and clear button. Preserve browser keyboard behavior; Escape may clear only when documented and must not trap focus.
• Announce result counts/no-results through a persistent aria-live="polite" status. Do not announce on every keystroke without debouncing.
• Keep focus in the input while typing. On explicit submission, move focus to a results heading only when that improves navigation; clearing returns to the input and restores the original archive.
• Use real result links and real pagination anchors. Mark current page with aria-current; give previous/next localized accessible names; maintain visible focus indication and adequate touch targets.
• New embedded searches should store state in a URL fragment to avoid sending terms to the static server while retaining share/back-forward behavior. The legacy /search?term= compatibility route necessarily sends its query and should document that static-host access logs may retain it.

I. Performance and Deployment

Expected static output is three JSON files such as /assets/search/library-en.json, library-es.json, and library-ar.json, one compiled client module, the generated HTML archives, and redirect-map entries. Load only the current locale index on first interaction; /search can load the selected locale and additional indexes only when requested. Cache fingerprinted client code immutably and JSON with normal deploy revalidation so new posts are visible after a release.

No database, persistent process, server-side search, external query service, environment variable, or production API is required. The result works on nginx, the current Nix environment, a Raspberry Pi serving static files, the existing VPS, and conventional static hosts. Build cost is a small pass over content and JSON serialization. Redirect behavior remains a deployment concern and must be validated against the selected host; Phase 6A changes no nginx/Nix/deployment file.

J. Proposed Phase 6B Implementation Checkpoints

6B.1 — Shared catalog and search assets

• Likely files: new src/utils/library-catalog.ts; JSON endpoint(s); LibrarySearch.astro and client script; existing Library components/layout; tests and README/report.
• Work: derive eligible locale catalogs, explicit sort, metadata-only index, normalization/ranking, locale messages, embedded search, and /search compatibility route. Replace inert English input; add Spanish/Arabic controls.
• Dependencies: approval of search fields, legacy route behavior, and noindex-record inclusion. No package dependency recommended.
• Acceptance: all search cases pass; JSON contains no drafts/body; current route count increases by one for /search; search remains noindex/out of sitemap; normal Library works without JavaScript.
• Validation: test:library, SEO/publication suites, both builds, route/link audits, browser keyboard/RTL/JS-disabled checks.

6B.2 — Library pagination and first-page redirects

• Likely files: catalog helper, route manifest/catchall dispatch, shared Library renderer/pager, redirect generator/map, tests/reports.
• Work: ten-item chunks, page 2+ descriptors, root page 1, invalid-page 404, six page-1 redirects, localized pagination. Remove bypassed listing snapshots from the three Library bodies only after rendered acceptance.
• Prerequisite: Phase 6C must supply the nine missing English records before /library/page/6 can pass preservation acceptance. If Phase 6C is not ready, stop this checkpoint with five English pages rather than fabricating content.
• Acceptance: every page has unique real content, explicit sort, valid canonicals/robots/sitemap state, no empty pages, and production URLs through current last pages.

6B.3 — Author model, archive roots, and byline links

• Likely files: author schema/records, catalog/route descriptors, new author archive presentation, BlogCard, BlogPostLayout, structured-data resolution helper, tests/report.
• Work: optional displayName, source-supported localized bios/headings, exact locale/slug resolution, archive pages/pagination, and shared linked bylines. Preserve BlogPosting attribution semantics.
• Dependencies: source reconciliation and human Spanish/Arabic review; approval of author-root hreflang and sitemap policy.
• Acceptance: roots and all nonempty numbered routes resolve; missing, duplicate, zero-post, and invalid authors fail/404 appropriately; no manual article arrays; no employment claims.

6B.4 — URL and SEO reconciliation

• Likely files: existing redirect generation/report outputs, publication tests, route-audit evidence; no hosting config until separately authorized.
• Work: add exact page-1/trailing normalization decisions, verify canonical, robots, sitemap, collisions, invalid boundaries, search noindex, and preserve all known paths.
• Acceptance: disposition matrix is executable and audited; no redirect to a generic Library root for a distinct page 2+ URL; staging safeguards remain.

6B.5 — Multilingual, accessibility, and performance acceptance

• Likely files: focused tests/report and only defect fixes in approved components.
• Work: English/Spanish/Arabic search, localized counts/dates, RTL pager, screen-reader announcements, keyboard/focus, JS-off browse, payload checks, mobile/desktop browser review.
• Acceptance: no fabricated translation routes; index sizes recorded; no external query transmission; complete build/audit matrix passes.

Each checkpoint must be an isolated reviewable diff. Failed validation stops the sequence. Phase 6C content changes should remain a separate checkpoint even if they are scheduled between 6B.1 and 6B.2.

K. Test Matrix

Area	Scenario	Test level	Expected result
Search	Empty/whitespace query	Unit + browser	Normal archive restored; no false zero-result state
Search	No match, one match, many matches	Unit + generated browser fixture	Correct deterministic count/order and polite announcement
Search	Case, repeated whitespace, punctuation	Unit	Normalized equivalent matches
Search	evaluación/evaluacion and Spanish punctuation	Unit + browser	Accent-folded key matches; original text displayed
Search	Arabic marked/unmarked query and punctuation	Unit + RTL browser	Diacritics/tatweel do not block intended match; direction stays RTL
Search	Clear/reset, back/forward, direct fragment	Browser	State and original page restore predictably
Search	Keyboard/focus/screen reader	Browser/manual assistive review	Native operation, visible focus, stable focus, live count
Search	JavaScript disabled	Generated HTML + browser	Paginated browse works; honest no-search message
Search	Draft/unpublished fixture	Unit + integration	Absent from JSON/results; body content never leaked
Search	Legacy term, q, type, valid/invalid offset	Integration + browser	Compatibility route safely normalizes and remains noindex
Pagination	First, middle, last page	Unit + generated HTML	Correct slices, count, prev/next/current links
Pagination	Page 1 aliases	Redirect test	Locale/author-specific 301 to root
Pagination	Beyond-last, zero, negative, noninteger	Route integration	No generated route; 404
Pagination	Zero posts / one page	Unit	No archive for zero-post author; no pager for one page
Pagination	Equal publication dates	Unit	Date descending then route ascending
Pagination	Add/remove/draft post	Fixture build	Page counts/routes/index update without manual edits
Author	Existing author with multiple posts	Unit + HTML	Localized record and correctly ordered posts
Author	New author / zero eligible posts	Fixture build	New route derives automatically only with eligible posts
Author	Missing/duplicate locale record	Unit + build failure	Explicit actionable validation error
Author	Invalid slug and page boundary	Route integration	404; no leakage
Author	Localized records and links	HTML + browser	Card/article links use matching locale archive
SEO	Canonical/robots/sitemap	Publication unit + both builds	Shared policy exactly matches generated descriptors
SEO	Hreflang	Unit + HTML	Roots only when explicitly equivalent; never page-number guessing
SEO	Staging versus indexing	Build integration	Global noindex/empty sitemap/no hreflang restored at end
Routing	Known production URLs/redirects	Route audit	Every matrix row matches route or redirect evidence
Quality	Internal links/assets	Existing audits	No broken links or missing images

L. Human-decision Register

Established project policy

• Static Astro, Markdown/MDX content collections, TypeScript, Zod, no database or headless CMS, no form/search backend, and no invented translations.
• Publication eligibility, canonicals, robots, sitemap, explicit translation relationships, staging protection, and external-canonical exceptions remain governed by Phases 3A/3B.
• Search, pagination, and author archives are required visitor capabilities; HubSpot internals need not be copied.

Codex recommendations requiring approval

1. Approve dependency-free generated metadata JSON search and defer full-body, fuzzy, and stemmed search until measured failures justify them.
2. Approve ten posts per page, unnumbered page-1 canonicals, and locale-specific 301 redirects from all six /page/1 aliases.
3. Approve /search as a preserved noindex compatibility page accepting term, q, legacy type, and offset; embedded Library search remains locale-only.
4. Approve indexable, self-canonical, sitemap-listed page 2+ and author archives, even though HubSpot omits those URLs from its sitemap and canonical markup.
5. Approve retaining accessible noindex/external-canonical articles in lists and internal search, matching current Astro behavior; they remain excluded from sitemap/hreflang and preserve their metadata.
6. Approve explicit author-root hreflang from author translationKey, with no hreflang for numbered pages.
7. Approve an optional localized displayName author field and a separate source-supported author-content reconciliation with human Spanish/Arabic review.
8. Approve scheduling the nine-post Phase 6C reconciliation before pagination and author page-6 acceptance. No empty or misleading page 6 should ship.

Not yet verified or intentionally deferred

• Production-cache timestamps do not prove origin freshness.
• Search query frequency and the value of body-only/fuzzy matching have not been measured; no analytics should be added to answer this without consent review.
• Exact hosting redirect syntax remains undecided and must be validated for the selected deployment target.
• Current author records are not adequate evidence for localized biography copy.
• Phase 6C publication/indexing/translation status for the nine missing posts is unknown, so the post-6C route projection is an upper-bound model.

No Phase 6B implementation, dependency installation, redirect, content edit, or deployment change is authorized by this proposal.