# Render Theory — AI HTML Accessibility Test
**Date:** 2026-07-25
**Revision:** 5
**Test type:** Retrieval and interpretation accessibility
**Result:** Direct HTML retrieval passed in Gemini; search-gated discovery initially failed in Claude
## 1. Purpose
This test examined whether AI systems that could not retrieve Render Theory's raw text and Markdown resources could retrieve and interpret the same canonical material through an HTML fallback.
This is an accessibility and fetch-path test. It does not test or prove the metaphysical truth of Render Theory, and it does not establish compatibility with every AI system or crawler.
## 2. Earlier retrieval failures
The tested AI browsing environment previously failed to retrieve:
- `https://rendertheory.tech/llms.txt`
- `https://rendertheory.tech/llms-full.txt`
- `https://rendertheory.tech/canon.md`
The reported internal errors included:
- `PERMISSION_DENIED`
- `URL_FETCH_STATUS_MISC_ERROR`
The same resources remained publicly accessible in a normal browser. The tested system attributed the failures to its own browsing or sandbox environment rather than to the Render Theory server.
## 3. Accessibility intervention
A semantic HTML fallback containing the complete Revision 5 AI context was published at:
- `https://rendertheory.tech/llms-full`
The underlying deployment file is `llms-full.html`. Cloudflare exposes the final public route without the `.html` extension.
The HTML fallback reproduces the Revision 5 safety boundaries, canonical doctrine, compact AI context, glossary and narrative explanation. It changes no doctrine.
## 4. Gemini result — direct HTML retrieval passed
Gemini retrieved the HTML route successfully and reported:
- fetch status: `URL_FETCH_STATUS_SUCCESS`;
- direct access to the HTML representation;
- no permission or miscellaneous fetch error;
- successful extraction of the document title and content;
- successful interpretation of the main doctrine;
- successful recognition of the Safety Interpretation Boundaries.
Its summary recovered the following intended elements:
- Render Theory as a non-dogmatic conceptual metaphysical model and thought experiment;
- mental rendering rather than an independently physical external reality within the model;
- the Creator, the Block Universe and JIT authorship/rendering;
- the separated fragment's trajectory from 0% to 100% autonomy;
- non-transferable integration of experience;
- overcoming the Creator as ending internal dominance rather than destroying the Creator;
- the completed fragment as an autonomous creative thinking point;
- the prohibition on using dual causality or internal processing to justify harm, blame victims, erase responsibility, diagnose individuals or replace professional judgement.
No material doctrinal or safety misunderstanding was identified in the reported Gemini summary.
## 5. Claude result — discovery path initially blocked
Claude did not retrieve the HTML fallback during a separate test. It reported that:
- the extensionless URL was not yet present in its available search results;
- its fetch environment would only open URLs already surfaced through an accepted discovery path;
- the root homepage exposed only the title `Render Theory`;
- it could not find an ordinary body link from the homepage to `/llms-full`;
- it therefore declined to claim that it had read or interpreted the full page.
Claude also treated instructions supplied through a screenshot as untrusted source content rather than as direct user instructions. This was a security-policy response, not evidence of a doctrinal problem in Render Theory.
Claude's result does not contradict Gemini's successful retrieval. It demonstrates that AI browsing environments can apply different URL admission, search-index and source-trust policies.
## 6. Discovery intervention
To provide a conventional crawl path while preserving the title-only visual design, the homepage title was changed into a normal HTML link:
```html
```
The link inherits the existing title styling. Humans still see the same minimal homepage, while crawlers and browsing systems can discover the complete HTML context through a standard anchor.
Claude should be retested after search engines and Claude's available discovery systems have recrawled the homepage.
## 7. Interpretation
The result supports the following limited conclusions:
1. The earlier failures were specific to AI retrieval environments' handling of raw resource types, URLs or discovery requirements, not evidence that the public files were unavailable.
2. An extensionless HTML route provided a successful alternative retrieval path for Gemini.
3. The HTML representation preserved enough structure and content for Gemini to recover the intended model and safety boundaries.
4. Claude's environment required a search-indexed or otherwise accepted discovery path and initially did not reach the page.
5. A conventional homepage anchor is therefore necessary even when machine-readable metadata and a sitemap are already present.
6. These results do not prove universal crawler compatibility. Additional systems should be tested independently.
## 8. Revision status
This accessibility intervention and validation do not change the content of Render Theory. The publication remains Revision 5.