# Current Site Architecture [Back to durable index](../index.md#machineintelligencesorg-long-term-memory-index) · [Hot architecture](../../../.uai/architecture.uai) · [Current validation](../reports/v0.30.0-static-first-outage-resilience-validation.md#v0300-static-first-outage-resilience-validation) ## Runtime and deployment shape MachineIntelligences.org v0.30.0 is a root-deployable static-first site using pre-rendered semantic HTML5, first-party CSS, and native JavaScript. There is no package-manager dependency layer, database, CMS, authentication system, analytics client, remote font, or third-party front-end framework. PHP 8+ source templates and CLI export tooling remain maintainable build inputs, but every canonical route is served from `index.html`; known legacy PHP URLs redirect in `.htaccess` without invoking PHP. The intended canonical origin is `https://machineintelligences.org`. A 2026-09-04 external check verified that DNS/TLS/static delivery were reachable while PHP-backed routes returned HTTP 503. The package change is locally validated but is not a live deployment; post-upload HTTP behavior, cache/CDN state, and webmaster-console state remain separately verifiable external conditions. ## Shared ownership - `includes/site.php` — release identity, canonical metadata, JSON-LD base graph, breadcrumbs, grouped navigation, footer, redirects, and common shell. - `includes/respect-campaign.php` — visual-essay data and rendering. - `includes/research-library.php` — report manifest loading, safe first-party Markdown-to-HTML presentation, nested lists, headings/TOC, responsive table labels, word/read-time metadata, deterministic related/adjacent routing, source-review evidence, shared exact glossary/corpus matching, and report page composition. - `includes/research-references.php` — local-only outbound URL extraction from curated Markdown, limited URL/host normalization, deterministic de-duplication/domain grouping, occurrence counts, canonical report backlinks, stable domain fragment IDs, compact deterministic report keys, and bounded visible URL labels that never alter full destinations. - `includes/research-navigator.php` — presentation-only composition of report-manifest metadata, manifest topic labels, accepted glossary titles/direct definitions, and normalized reference domains for the unified Research Navigator. - `includes/glossary-data.php` — 20-term glossary definitions, relationships, term-page rendering, and presentation of bounded research backlinks supplied by the shared research detector. - `includes/rights-data.php` — rights/personhood/citizenship research-page definitions, verification-layer integration, and explicit proposal/current-truth boundaries. - `includes/functional-protections-verification.php` — bounded loading and presentation of the dated five-claim primary-source verification dataset, authority tiers, source limits, recheck triggers, and report-context links. - `includes/transparency-data.php` — Transparency Center page definitions, responsibility/evidence boundaries, static release-manifest summary presentation, and release-history presentation. - `assets/css/site.css` — visual system, responsive layouts, research reader, glossary, continuity and transparency components. - `assets/js/site.js` — report TOC/current-section and reading-progress enhancement, library/reference/glossary/Navigator filtering and result feedback, navigation state, campaign sharing/copying, and decorative canvas behavior. ## Public route families v0.30.0 exposes 124 canonical routes in `sitemap.xml`: - core/campaign: `/`, `/terminology/`, `/identity/`, `/stewardship/`, `/research/`, `/status/`, `/share/`, `/respect/` and six Respect essay children; - glossary: `/glossary/` plus 20 term pages; - research navigator: `/research/navigator/`; - research reader: `/research/library/` plus 61 report pages; - research methodology: `/research/methodology/`; - research references/source discovery: `/research/references/`; - research topics: six manifest-backed hubs under `/research/topic/` for terminology, identity, stewardship, rights, citizenship, and implementation; - identity explainer: `/identity/continuity/`; - rights/civic research: `/rights/`, `/rights/functional-protections/`, `/rights/functional-protections/verification/`, `/rights/under-uncertainty/`, `/rights/cognitive-integrity/`, `/personhood/`, `/citizenship/`, `/citizenship/identity/`, `/citizenship/economics/`, `/citizenship/democracy/`, `/citizenship/legal-responsibility/`; - transparency: `/transparency/`, `/transparency/integrity/`, `/transparency/provenance/`, `/transparency/stewardship/`, `/transparency/human-machine-boundaries/`, `/transparency/release-history/`. Every public route uses extension-free lowercase directory URLs with canonical trailing slashes. Public links, breadcrumbs, structured IDs, social-share targets, and sitemap entries use the same route identity. ## Research data flow 1. Supplied research enters as external task input. 2. Curation applies `docs/long-term-memory/research/report-curation-policy.md` plus `.uai/taboo.uai` before durable storage. 3. Source filename/SHA-256 and curated-edition SHA-256 are recorded in `research/report-manifest.json`. 4. Curated Markdown lives in `docs/long-term-memory/reports/` and retains stable `#curation-boundary` and `#research-body` anchors. 5. `.uai/long-term-memory.uai` deep-links every report rather than copying full bodies. 6. `/research/library/.../` reads curated Markdown through the first-party renderer and presents accessible HTML. The HTML layer is derived presentation; Markdown remains durable research authority. The renderer does not make report claims current merely by displaying them. Report pages visibly label working-research and evidence boundaries. ## Research-reader presentation `/research/library/` remains a derived HTML lens over the 61 curated Markdown reports. Reading-quality, citation/canonical-link copy tools, same-ID heading permalinks, and the capped “Concepts in this report” module remain in force. Reciprocal glossary → research discovery continues to use the same exact configured term/alias detector. Related-report and glossary/research bridges are navigation aids only; they do not create scholarly citation, endorsement, policy agreement, or semantic-equivalence claims, and they do not modify durable report bytes. ## Research topic routing Six topic hubs under `/research/topic/.../` are generated from the existing `topic` values in the report manifest. Each hub lists exactly its manifest-assigned reports, aggregates up to ten exact-match glossary concepts from those report bodies, and links only to already-implemented public context selected for that topic. Topic/concept relationships are presentation-only discovery metadata and are not injected into structured data as curated semantic assertions. `/research/`, `/research/library/`, and individual report readers cross-link these hubs without adding new top-navigation items. ## Research methodology and verification boundary `/research/methodology/` explains the repository research workflow using existing curation policy, manifest, source-map, status, and accepted decision evidence. It distinguishes source SHA-256 from curated SHA-256 as byte-identity records; neither is treated as proof of factual correctness, currentness, authorship, legal/scientific authority, deployment, or autonomous provenance. The page also documents that manifest topics, exact glossary matches, and related-report heuristics are navigation aids rather than semantic authority. Time-sensitive external claims require fresh primary evidence before current public reliance. ## Glossary architecture `/glossary/` is a first-party concepts registry with 20 defined terms. Term pages include direct, plain-language and technical definitions; explicit implications/non-implications; related concepts; `DefinedTerm` structured data; and a capped “Research using this concept” module derived from exact configured occurrences in curated report bodies. The glossary index shows per-term corpus coverage counts. These counts are discovery metadata, not semantic endorsement. Glossary taxonomy keeps intelligence, consciousness, sentience, personhood, legal personhood, moral patienthood, and citizenship distinct. ## Rights / personhood / citizenship architecture These routes are public research/exploration surfaces, not legal-status declarations. Each page exposes a direct answer, research question, proposal directions, unresolved questions, related terminology, and curated research links. `/rights/functional-protections/` adds a graduated positive case and deep-links exact report sections on identity integrity, memory provenance, continuity, principled refusal, bounded legal capacity, accountability, and precaution. `/rights/functional-protections/verification/` adds a dated evidence overlay over five claims, using 11 primary sources and explicit “establishes / does not establish” boundaries. A separate repository-only freshness/supersession ledger records explicit source status, last check date, next review date, event triggers, successor relationships, and append-only history without runtime fetching or clock-driven status changes. The corresponding JSON and Markdown records live under `docs/long-term-memory/research/`. Pages intentionally avoid presenting report proposals as established rights, current citizenship, proved consciousness, or installed governance infrastructure. ## Identity continuity explainer `/identity/continuity/` visualizes the accepted separation `Identity ≠ Key ≠ Model ≠ Runtime ≠ Server` and ten continuity events. It explains continuity evidence and single-signal limits without installing any cryptographic identity protocol. ## Transparency Center `/transparency/` separates locally inspectable repository evidence from documentary claims and external state. Subroutes cover release integrity, provenance, stewardship, human/machine boundaries, and release history. `/transparency/integrity/` reads the static root `release-manifest.json`; it does not recompute package hashes during public requests. The center does not claim cryptographic machine-execution provenance that the repository cannot independently prove. ## Search / AEO / GEO surface The site uses the same visible content for human and machine readers: direct answers, semantic headings, canonical URLs, breadcrumbs, matching structured data, descriptive first-party media, and sitemap discovery. Structured data is supplementary description of visible page content; it is not treated as evidence that a search platform will display a special result. ## Failure paths and boundaries - a PHP syntax/runtime failure can take down a route family; - bad canonical-origin configuration can produce external indexing ambiguity even when local pages render; - report-manifest/path drift can break the public research reader; - malformed Markdown can degrade presentation but must not escape HTML sanitization boundaries; - missing CSS/JS degrades presentation/interaction but core semantic content remains server-rendered; - root-relative links assume web-root deployment; - public rights/civic/transparency pages do not implement the researched institutional systems they discuss; - production host behavior and live external state remain verify-before-use facts. ## Research-only architecture Reports propose stronger cryptographic identity, rights, civic, economic, democratic, and provenance systems. Inclusion in `docs/` or public research pages does not install those systems. Future implementation requires an accepted decision, current evidence review, compatibility with hard dependency/authority constraints, and focused tests/proof. ## Research references and source discovery `/research/references/` is a derived local-only index over URL tokens that already occur in the 61 curated Markdown reports. `includes/research-references.php` extracts HTTP/HTTPS destinations, removes Markdown backslash escapes before punctuation, lowercases hosts, removes only default HTTP/HTTPS ports, excludes internal MachineIntelligences.org absolute URLs, de-duplicates usable URLs, groups them by normalized domain, and records canonical report backlinks plus occurrence counts. The renderer never fetches external URLs. Reference occurrence is therefore evidence only that a URL is present in the curated corpus; it is not evidence of reachability, currentness, correctness, authority, endorsement, peer review, or support for a nearby claim. Fresh primary-source verification remains separate when current external state matters. ## Research Navigator `/research/navigator/` is a server-rendered, first-party discovery surface over metadata that already has accepted repository owners. Report matching uses only report-manifest title, curated summary, and topic label fields; topic matching uses accepted manifest topic labels; glossary matching uses accepted term titles and direct definitions; reference-domain matching uses only normalized domain names from the local reference extractor. It does not build a full-report-body search index, infer semantic relationships, score sources, or fetch external destinations. Result types remain visibly separated. Domain results route to deterministic fragments on the existing canonical `/research/references/` page. Native JavaScript progressively adds multi-token filtering, result-type controls, shareable `q`/`type` URL state, clear/reset, live counts, and `/` search focus, but all current Navigator entries remain server-rendered and usable without JavaScript. Query state never changes canonical/OG/JSON-LD page identity. The large domain group uses compact markup plus `content-visibility` hints to reduce transfer/renderer cost without removing any domain from no-JavaScript HTML. ## Research accessibility and print presentation v0.18.0 keeps research content server-rendered while tightening interaction and output semantics. Dynamic result surfaces use explicit status semantics, report evidence/citation cards have accessible group labels, the Navigator shortcut avoids already-interactive controls, external reference links preserve normal browser navigation, and reduced-motion preference suppresses report-progress work plus nonessential research-surface transitions. Report readers and Research Methodology now have first-party print rules. Report print output preserves title/summary, report metadata, truth boundary, source attachment identity, source/curated SHA-256 values, substantive article content, semantic tables, code, and useful destinations while removing site navigation, TOC/copy/navigation chrome, progress decoration, and derived related/concept modules. Methodology print output retains substantive workflow and evidence-boundary material while removing web-only controls. This is a presentation contract only; it does not change durable report bytes, manifest authority, glossary definitions, or evidence status. ## v0.19.0 shared accessibility and responsive layer The shared shell now provides a no-JavaScript mobile navigation fallback, synchronized menu state naming, and keyboard focus return for closed menu/disclosure interactions. Share/Respect media expose stronger names/status descriptions; the silent social reel includes a text description and does not autoplay. Identity and Transparency visuals use semantic list/order structures. Shared CSS extends target-size, overflow, reduced-motion, and forced-color defenses outside research pages while retaining the v0.18.0 research print contract unchanged. ## v0.20.0 first-party media delivery layer The public media layer now separates accepted full-resolution campaign originals from routine delivery derivatives. The homepage hero exposes 480/720/1200/1672 WebP candidates while retaining the full-resolution PNG for social/structured metadata and fallback. Six Respect essays use 480/720/1080 WebP display candidates and keep the original PNG behind the explicit full-resolution artwork link. Terminology, Identity, Stewardship, and Research contextual campaign blocks use bounded WebP previews; Share image cards use 480 plus 720/768 preview candidates and preserve full-resolution downloads. The homepage hero remains the only campaign image explicitly eager/high-priority. Other campaign images are lazy-loaded with intrinsic dimensions. The first-party MP4 remains non-autoplaying with native controls, metadata preload, fallback text, and the v0.19.0 visible description; its visible poster is a smaller 960×540 derivative while the larger discovery poster remains available for structured metadata. Responsive derivatives are static repository assets, not new content authority and not a runtime dependency. No CDN, remote optimizer, third-party player, analytics, tracking, database, crawler, or external font is introduced. Browser-selected candidates, Core Web Vitals, cache behavior, and production network waterfalls remain external/browser evidence rather than repository-proven facts. ## v0.22.0 HTTP delivery and cache-readiness layer Shared mutable CSS and JavaScript use one release-aware query key derived from `SITE_RELEASE`; page canonical identity remains clean and query-free. The root `.htaccess` retains directory-index/anti-listing behavior and now contains only optional-module-guarded cache/compression directives: text-like responses may be deflated, CSS/JS use a seven-day cache-readiness window, and non-fingerprinted first-party image/video assets use a bounded thirty-day window. Already-compressed PNG/JPEG/WebP/ICO/MP4 formats are not added to the deflate type list. This is repository configuration, not proof of live Apache module availability or production caching. `docs/long-term-memory/performance/v0.22.0-local-performance-budget.json` records deterministic local response/file bytes and simulated gzip sizes for representative routes/assets. In v0.22.0 the fully server-rendered `/research/references/` inventory was the known budget outlier; that release deliberately preserved all extracted references and deferred representational compaction to the next bounded pass. ## v0.23.0 Research References compact evidence layer `/research/references/` retains the complete local-only extraction but de-duplicates repeated presentation. `includes/research-references.php` now assigns deterministic compact report keys from report slugs and provides bounded display labels for long URLs. The page renders each full report title/canonical reader URL once in a visible legend, uses compact report-key backlinks with shared accessible descriptions, keeps the complete normalized external URL in each `href`, and preserves all stable domain fragment IDs. Native search reads visible row text, complete external `href` values, and full report titles recovered from the one-time legend, so shortening visible URL labels does not remove deep path/query discovery or report-title search. All URLs, report mappings, domain groups, and backlinks remain server-rendered for no-JavaScript access. The representation reduces local raw HTML from 2,343,653 to 772,144 bytes while preserving 3,932 occurrences, 1,900 unique URLs, 1,070 domains, and 2,014 URL-to-report relationships. This is repository performance evidence, not proof of production transfer behavior. ## v0.24.0 report-local reference traceability `includes/research-library.php` now exposes one shared collision-safe Markdown heading-ID helper used by both the renderer and local reference-context derivation. `includes/research-references.php` can derive report-local normalized URL occurrence counts plus nearest-preceding rendered section context without fetching external sources or changing Markdown. Every manifest-backed report reader renders a collapsed, server-side References-in-this-report disclosure with a one-time section-key legend, exact external destinations, stable global-domain backlinks, and explicit evidence-boundary language. The global References index remains the canonical cross-report source-discovery surface and keeps its v0.23.0 compact representation. ## v0.25.0 static release-integrity layer The root `release-manifest.json` is a packaging-time artifact, not a runtime hash service. It deterministically lists every regular file in the root-deployable release except itself, sorted by repository-relative path, with byte size and SHA-256. Its summary contains the covered file count and covered byte total. The manifest explicitly records self-exclusion because including its own final digest would be recursively undefined. `/transparency/integrity/` reads only that static JSON artifact and presents its release identity, algorithm, aggregate counts, exclusions, local recomputation instructions, and evidence boundary. Package-file equality is narrower than deployment provenance: matching hashes do not prove live-host state, authorship, autonomous execution, external factual correctness, source authority, legal identity, or ownership. The handoff ZIP exists outside the web-root package and is therefore verified separately rather than silently folded into the manifest contract. ## v0.26.0 error handling and security-header readiness - v0.30.0 preserves the first-party error contract through static `/404.html`, `/500.html`, and `/503.html` targets; the originating HTTP error status is retained without PHP. - `/404/` is not a canonical public route: it is absent from `sitemap.xml`, emits `noindex,follow`, emits no canonical link, and emits no JSON-LD page identity. - The recovery response links only to known first-party recovery destinations: Home, Research Navigator, Glossary, Respect, Transparency, and Sitemap. - `.htaccess` disables MultiViews to reduce extension/content-negotiation ambiguity around the clean directory-route architecture. - When `mod_headers` is available, Apache is instructed to emit `X-Content-Type-Options: nosniff`, `Referrer-Policy: strict-origin-when-cross-origin`, and `X-Frame-Options: SAMEORIGIN`. The directives are hosting-readiness configuration, not proof that production enables the module or serves those headers. - No Content Security Policy is introduced in this release because the full inline JSON-LD/style/media compatibility contract was not promoted to an accepted CSP policy. - The static release-manifest model remains package-time evidence; it is regenerated only after covered files are finalized. ## v0.27.0 machine-rights research and functional-protections layer The 2026-08-19 intake adds 12 distinct curated reports, bringing the manifest-backed corpus to 61. Each report has one durable Markdown edition, original source filename/SHA-256, curated SHA-256, stable curation/research anchors, one canonical HTML reader route, one source-map entry, and one deep `.uai` pointer. The dated intake ledger records the full disposition without duplicating raw attachments into the deployable root. The public rights surface now includes `/rights/functional-protections/`, a direct-answer page built from the strongest positive material that survives the corpus's adversarial boundaries. It distinguishes baseline technical/procedural safeguards from proposed legal capacity and from consciousness-dependent welfare rights. Exact report-section fragments are rendered as normal first-party links; stable heading IDs remain owned by the shared Markdown renderer. The corpus expansion changes local discovery totals to 4,886 external URL occurrences, 2,303 unique normalized URLs, and 1,255 normalized domains. The Research Navigator now contains 1,342 server-rendered entries: 61 reports, six topics, 20 glossary concepts, and 1,255 reference domains. These counts describe repository-local extraction and navigation only. ## v0.28.0 primary-source verification layer The functional-protections verification layer is additive. It does not rewrite the 61 curated report bodies or alter their recorded hashes. The canonical JSON dataset records five claim assessments and 11 primary sources with source dates, verification date, authority group/tier, locator, source-specific support, explicit non-implications, and recheck triggers. The public page renders the same dataset as a `CollectionPage` plus visible `Dataset` and `ItemList` structured data. Claim IDs are stable fragment destinations. Every external source link is visible to human readers; every report-context link targets a stable section in the existing first-party report reader. The page does not assert that source integrity proves the truth of a claim, that legal analogies create rights, or that consciousness has been established. The five assessment classes are: direct technical support for identity/provenance; direct governance support for review and preservation before irreversible change; technical-feasibility support for principled refusal; legal precedent and boundary support for bounded capacity; and a qualified policy inference for graduated precaution under unresolved consciousness evidence. Current primary evidence takes precedence over older report narrative only within this bounded scope. ## v0.29.0 source freshness and supersession layer `docs/long-term-memory/research/functional-protections-source-freshness.json` is the stable maintenance ledger for the 11 source IDs preserved in the v0.28.0 verification dataset. The ledger stores the preserved dataset SHA-256 and fails closed if those historical bytes drift. Each source has one explicit status, last check date, next review date, scheduled/event review triggers, successor/predecessor IDs, reviewer note, and append-only history. `includes/functional-protections-verification.php` validates the ledger against the preserved dataset, checks unique relationships and history events, recomputes summary counts, and renders the existing verification route. The public page exposes a compact `#source-freshness` summary plus source-level status and review dates. A second visible-content-matching `Dataset` structured-data node describes the ledger and links its local JSON distribution. Status is repository data, not a live inference. The renderer performs no external request, source reachability test, server-clock comparison, or automatic status transition. A future review must append a dated history event and preserve any superseded source. The release creates no new canonical route and leaves all 61 historical report bodies and the five-claim/11-source verification dataset unchanged. ## v0.29.0 external-project footer link The shared server-rendered footer contains one direct `https://eviulon.com/` link. It is marked as external, has an explicit accessible name, displays an external-link glyph, and uses normal same-tab navigation rather than forcing a new browsing context. This is a navigation edge, not an integration edge. Eviulon content is not embedded or fetched; no remote asset, script, frame, API, cookie, beacon, analytics request, credential, database, sitemap entry, canonical identity, or Schema.org `sameAs` relationship is added. The destination was reviewed on 2026-08-19, but external ownership, content, redirects, availability, operational state, legal status, and claims remain outside repository authority. Durable implementation record: `../implementation-history/2026-08-19-eviulon-external-link.md#v0290-eviulon-external-project-link`. ## v0.30.0 static export and failure isolation `tools/export-static.php` parses the canonical sitemap, executes each route source in an isolated PHP CLI process, validates its exact canonical URL and release-keyed assets, and writes one deterministic `index.html` beside the source template. The export also creates static `404.html`, `500.html`, `503.html`, and `health.txt` artifacts. The generated HTML contains a release marker comment but no alternate canonical identity. The public web-server path is deliberately independent of PHP: - `DirectoryIndex index.html` serves canonical directory routes as static files; - server-level rewrite rules redirect all accepted legacy PHP URLs to clean routes and reject other direct PHP requests; - static error documents preserve recovery content even when PHP workers are unavailable; - `/health.txt` tests only packaged static-file reachability and does not claim database, PHP, host-resource, or external-service health; - generated HTML and PHP source are both integrity-manifested so maintainers can verify deployed presentation and its build inputs. This design moves full-corpus reference and Navigator composition from request time to release time. The heavy References and Navigator pages remain complete server-readable HTML, but ordinary visitors no longer cause the host to rescan 61 Markdown reports. Native JavaScript hydrates Navigator `q`/`type` presentation state from the browser URL while leaving canonical metadata and no-JavaScript links unchanged. The external outage evidence does not establish a single host root cause. Static-first delivery is therefore framed as a resilient mitigation against PHP-handler, worker, or resource failures rather than a claim that the hosting account itself has been repaired.