<a id="v0230-research-references-scalability-validation"></a>
# v0.23.0 Research References Scalability Validation

Date: 2026-08-12  
Release: 0.23.0

## Scope

This validation covers the v0.23.0 Research References transfer-size and interaction-scalability pass while preserving the accepted v0.22.0 delivery/cache contract, the 109-route public surface, the 49-report research authority, the 42-entry source-bundle disposition ledger, responsive media, accessibility/print behavior, clean URLs, and UAIX/taboo boundaries.

## Compact References evidence checks

The local extractor recomputes exactly:

- **49** curated reports;
- **3,932** external URL occurrences;
- **1,900** unique normalized external URLs;
- **1,070** normalized domain groups;
- **2,014** unique normalized-URL-to-report backlink relationships.

The rendered `/research/references/` page was checked against that extractor:

- all 1,900 unique normalized URLs appear exactly once as the primary external `href` for a reference row;
- every row's occurrence count matches the extractor;
- every compact report backlink set matches the extractor's canonical report-reader relationship set exactly;
- all 1,070 `reference-domain-*` fragment IDs match the shared deterministic domain-ID helper;
- the visible report-key legend contains exactly 49 deterministic, collision-free keys;
- each key maps one-to-one to the complete visible report title and canonical clean report-reader URL;
- every compact backlink's shared `aria-describedby` target resolves to that visible full-title mapping;
- the page remains fully server-rendered: no JavaScript is required to reach external URLs, report readers, report-key mappings, or domain groups.

Native JavaScript source was checked to ensure reference filtering includes the complete external anchor `href` and full report titles recovered from the one-time legend in addition to visible row text. Therefore visual URL shortening does not remove deep path/query searchability or report-title search.

External source occurrence remains a local corpus fact only. No external link was fetched or scored by the site runtime during this validation.

## Deterministic size comparison

v0.22.0 baseline for `/research/references/`:

- raw HTML: **2,343,653 bytes**;
- deterministic gzip-9 simulation: **172,945 bytes**.

v0.23.0 result:

- raw HTML: **772,144 bytes**;
- deterministic gzip-9 simulation: **108,967 bytes**.

Reduction:

- raw HTML: **1,571,509 bytes / 67.1%**;
- gzip simulation: **63,978 bytes / 37.0%**.

The route is now inside the existing **1,048,576-byte** advisory raw-response budget. The deterministic gzip number is comparison evidence only; it is not observed production transfer size.

Machine-readable inventory: `../performance/v0.23.0-local-performance-budget.json`  
Human-readable inventory: `../performance/v0.23.0-local-performance-budget.md#v0230-local-performance-budget`

## Public route, metadata, and delivery regression

The canonical sitemap retains **109** clean public routes, **8** image entries, and **1** video entry.

All 109 canonical routes were locally required to provide:

- HTTP 200;
- exactly one H1, one main landmark, and one footer;
- non-empty unique title and meta description;
- expected self-canonical under `https://machineintelligences.org`;
- matching Open Graph page URL;
- parseable JSON-LD;
- unique IDs and resolving ARIA ID references;
- named buttons and image alternative attributes;
- exactly one `/assets/css/site.css?v=0.23.0` reference and one deferred `/assets/js/site.js?v=0.23.0` reference;
- no release query key in canonical or Open Graph page identity;
- no public `.php` link targets.

The v0.22.0 optional Apache delivery configuration remains intact: deflate/expires/headers behavior is guarded with `<IfModule>`, and compressed binary image/video formats are not added to the deflate type list. Production module availability remains unverified.

Clean-URL regression coverage remains **230 redirect assertions**: 108 missing-trailing-slash variants, 108 clean-route `/index.php` variants, root `/index.php`, and 13 retained legacy root PHP endpoints.

## Local links, media, and accessibility regression

- **3,128** unique rendered local path/fragment target combinations were checked from canonical page output.
- Responsive image markup exposes **42** `srcset` candidates; every candidate exists and its declared width matches the actual local image width.
- Rendered local image/video/poster references resolve to repository files.
- The native social video remains non-autoplaying.
- Reference domain disclosures retain at least the accepted 48-pixel minimum summary height, stable hash targeting, and reduced-motion handling.
- The compact report-key system uses shared visible descriptions instead of repeated hidden full report titles.

## Research authority preservation

No research body or accepted glossary definition changed in this task.

- **49/49** curated report SHA-256 values match `report-manifest.json`.
- **49/49** original source SHA-256 values were re-matched against available supplied task evidence by hash.
- Every curated report retains `#curation-boundary` and `#research-body`.
- Every one of the 49 manifest-backed report files is byte-identical to the accepted v0.22.0 package.
- `report-manifest.json`, `includes/glossary-data.php`, and `source-bundle-intake-2026-08-12.json` are byte-identical to v0.22.0.
- The source-bundle ledger remains exactly **42 entries: 24 new-curated + 18 duplicate-existing**.

## Syntax, documentation, UAIX, and hygiene

- PHP files: **130**; all pass `php -l`.
- `assets/js/site.js` passes `node --check`.
- **313** repository-local Markdown links and **441** explicit UAIX repository-path references are rechecked during package validation.
- The required UAIX setup/update URL remains present exactly once in each required owning file.
- `.uai` package/current-state records advance to **0.23.0** and retain the 49 report deep pointers.
- Active handoff buckets remain inert placeholders.
- TABOO-001 containment passes without copying its prohibited literal into this report.
- Lightweight credential-marker hygiene passes after excluding ordinary research prose that merely contains hyphenated natural-language substrings.

## Evidence limitations

This is repository/local-server validation. It does **not** establish external reference reachability, source correctness/authority/currentness, production Apache modules, HTTP compression, cache behavior, browser/network timing, Core Web Vitals, live DNS/TLS/redirect state, indexing, webmaster-console state, accessibility certification, or independently verifiable machine-execution provenance.

No live deployment was performed.

## Next bounded action

Add per-report reference context using the existing local extractor: derive each report's normalized external references and, where deterministic, nearest existing report-section anchor; expose that as a bounded server-rendered report-reader module without editing authoritative Markdown or fetching external sources. The full task is stored in `.uai/next-recursive-prompt.uai`.
