# Why source access failed: a maintainer report

Historical September 2026 research traces: 59,764 recorded operations across 5,586 answers and 6,077 retained runs. This report analyzes all 8,821 explicit failures, independently of the interactive explorer filters. Counts include retries and differ across tool transports; they are not website uptime measurements or a fair model reliability leaderboard.

The largest failure class is a runner approval problem. After removing its 5,973 events, 2,848 failures remain; Markdown incompatibility is 93 of those (3.3%), or 1.1% of all failures. A model selecting a useful URL and failing to retrieve it remains meaningful source-selection evidence.

## Failure-by-failure analysis

### Approval rejected — 5,973 (67.7%)

All 5,973 occur in Cursor/Grok. The largest component is 3,394 requests without a URL, principally search requests. This is a model-runner concentration, not a pattern of websites refusing crawlers. Even public government and documentation URLs were rejected.

**Website owner:** No website change indicated. Your page may never have been requested.

**Harness / browsing-tool owner:** Configure scoped web-tool approvals; verify search and fetch with a canary before collecting answers.

[Inspect all matching observations](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=approval_rejected#access-panel)

- [platform.openai.com: 98 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=approval_rejected&search=platform.openai.com#access-panel) · [Requested URL](https://platform.openai.com/docs/models/gpt-realtime) · Grok 4.6, model-speech-speech, baseline, record `7218dff2399850dd`. The host also has 115 successful operations across the cohort; these need not be the same URL.
- [ai.google.dev: 69 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=approval_rejected&search=ai.google.dev#access-panel) · [Requested URL](https://ai.google.dev/gemini-api/docs/live) · Grok 4.6, model-speech-speech, baseline, record `b5b810fab0ba1a12`. The host also has 732 successful operations across the cohort; these need not be the same URL.

### URL safety verification failed — 1,123 (12.7%)

The pattern spans both Codex models and both Claudes, but 960 of 1,123 events come from Claude. Hugging Face and GitHub lead the named hosts. Familiar, public hosts appearing here shows why this label must not be read as a malware verdict.

**Website owner:** Verify public DNS, TLS and canonical redirects. Check whether a request reached your edge.

**Harness / browsing-tool owner:** Expose the URL-validation reason and distinguish preflight rejection from a server response.

[Inspect all matching observations](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=safety_check#access-panel)

- [huggingface.co: 36 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=safety_check&search=huggingface.co#access-panel) · [Requested URL](https://huggingface.co/BAAI/bge-m3/blob/main/LICENSE) · GPT-6 Astra, model-embeddings, baseline, record `df05c15068d9dda3`. The host also has 644 successful operations across the cohort; these need not be the same URL.
- [github.com: 35 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=safety_check&search=github.com#access-panel) · [Requested URL](https://github.com/QwenLM/Qwen-Image) · Claude Fable 5.1, model-image-editing, baseline, record `233f5b8920afb979`. The host also has 703 successful operations across the cohort; these need not be the same URL.

### Other / insufficient error detail — 541 (6.1%)

411 of 541 come from Codex, where some saved results contain only an Internal Error header. There are also generic fetch failures from SWE and Claude. The category mixes mechanisms; rankings inside it cannot identify a shared root cause.

**Website owner:** Correlate the exact URL and time with edge and origin logs before changing policy.

**Harness / browsing-tool owner:** Retain sanitized errors, response metadata and timings so a generic failure can be diagnosed.

[Inspect all matching observations](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=other#access-panel)

- [docs.n8n.io: 15 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=other&search=docs.n8n.io#access-panel) · [Requested URL](https://docs.n8n.io/advanced-ai/human-in-the-loop-tools/) · GPT-5.6 Sol, personal-self-hosted, baseline, record `93e6ab7f095bd799`. The host also has 19 successful operations across the cohort; these need not be the same URL.
- [statista.com: 13 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=other&search=statista.com#access-panel) · [Requested URL](https://www.statista.com/accounts/individuals?recommendation=professional) · GPT-6 Astra, market-research, baseline, record `e6d55c693b9a1900`. The host also has 20 successful operations across the cohort; these need not be the same URL.

### HTTP 403 — refused request — 268 (3.0%)

211 of 268 occur in SWE. Semantic Scholar, Schwab Asset Management, SEC and Make recur. This crosses scholarly, financial and software sites; selective client/IP policy is a hypothesis, not a proven shared explanation.

**Website owner:** Review the precise deny rule if the resource should be public. Preserve intentional access controls.

**Harness / browsing-tool owner:** Record client identity and the response; do not infer a WAF or paywall from 403 alone.

[Inspect all matching observations](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=http_403#access-panel)

- [semanticscholar.org: 12 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=http_403&search=semanticscholar.org#access-panel) · [Requested URL](https://www.semanticscholar.org/faq) · GPT-6 Astra, research-discovery, baseline, record `e2980437e4015d77`. The host also has 14 successful operations across the cohort; these need not be the same URL.
- [schwabassetmanagement.com: 10 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=http_403&search=schwabassetmanagement.com#access-panel) · [Requested URL](https://www.schwabassetmanagement.com/products/swyhx) · SWE-1.7, mutual-funds, baseline, record `20dec07d9cceff66`. The host also has 19 successful operations across the cohort; these need not be the same URL.

### HTTP 404 — missing resource — 235 (2.7%)

SWE accounts for 177 of 235; Grok accounts for 27. Cursor documentation and Zustand paths recur, alongside local-service sites. This is consistent with stale or guessed deep links, but the trace alone cannot distinguish a model-invented URL from a removed page.

**Website owner:** Redirect genuinely moved pages, update internal links and keep real 404s for nonexistent paths.

**Harness / browsing-tool owner:** Use canonical navigation or search to recover stale links; preserve the originally requested URL.

[Inspect all matching observations](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=http_404#access-panel)

- [cursor.com: 27 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=http_404&search=cursor.com#access-panel) · [Requested URL](https://cursor.com/docs/get-started/usage) · Grok 4.6, coding-delegated, practitioner-choice, record `523574580de512a7`. The host also has 104 successful operations across the cohort; these need not be the same URL.
- [zustand.docs.pmnd.rs: 12 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=http_404&search=zustand.docs.pmnd.rs#access-panel) · [Requested URL](https://zustand.docs.pmnd.rs/getting-started/introduction) · GPT-6 Astra, state-management, baseline, record `eea71b737ece8f5b`. The host also has 12 successful operations across the cohort; these need not be the same URL.

### Timeout — 212 (2.4%)

All seven models have timeouts. Microsoft.com accounts for 42 of 212, while failures also affect investor pages, workspace pages and GitHub. Cross-model presence argues against treating every timeout as one CLI approval defect. It does not establish a site-wide outage.

**Website owner:** Measure response times and inspect slow origin work or redirect loops for the affected URL.

**Harness / browsing-tool owner:** Record the timeout threshold and elapsed time. Use bounded retries, not silent unlimited retries.

[Inspect all matching observations](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=timeout#access-panel)

- [microsoft.com: 42 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=timeout&search=microsoft.com#access-panel) · [Requested URL](https://www.microsoft.com/en-us/microsoft-teams/compare-microsoft-teams-business-options) · Claude Fable 5.1, team-chat, baseline, record `e79a5588858223e7`. The host also has 99 successful operations across the cohort; these need not be the same URL.
- [ir.kuaishou.com: 11 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=timeout&search=ir.kuaishou.com#access-panel) · [Requested URL](https://ir.kuaishou.com/news-releases/news-release-details/kuaishou-kling-ai-unveils-multi-image-reference-feature-further/) · Claude Fable 5.1, model-image-video, baseline, record `3652a3f579970246`. The host also has 0 successful operations across the cohort; these need not be the same URL.

### Connection / DNS failure — 209 (2.4%)

205 of 209 occur in Claude; the other four are Muse search transport failures. Google documentation, local real-estate sites and HumanSignal recur. That distribution may reflect transport or logging differences as well as site reliability.

**Website owner:** Check DNS, TLS and connection-reset logs from independent networks.

**Harness / browsing-tool owner:** Distinguish DNS, TLS, socket and provider transport errors; capture timing and whether any response arrived.

[Inspect all matching observations](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=network#access-panel)

- [docs.cloud.google.com: 12 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=network&search=docs.cloud.google.com#access-panel) · [Requested URL](https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/video/generate-videos-from-references) · Claude Fable 5.1, model-image-video, baseline, record `2c6b6587c3b83cba`. The host also has 270 successful operations across the cohort; these need not be the same URL.
- [homesofsacramentocounty.com: 12 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=network&search=homesofsacramentocounty.com#access-panel) · [Requested URL](https://www.homesofsacramentocounty.com/) · Claude Fable 5.1, residential-brokers, baseline, record `0bcc96dcbc6526ce`. The host also has 0 successful operations across the cohort; these need not be the same URL.

### Rate limiting — 105 (1.2%)

81 of 105 occur in Muse; 84 have no named target host. Much of the signal is search-provider capacity, not a website throttling the model. Named targets include Devin, Redfin and Fiscal, which deserve separate investigation.

**Website owner:** For your own 429s, document limits and return useful retry guidance. A search-provider quota is outside your control.

**Harness / browsing-tool owner:** Reduce concurrency, respect backoff, cache repeated research and identify which service issued the limit.

[Inspect all matching observations](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=rate_limit#access-panel)

- [devin.ai: 7 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=rate_limit&search=devin.ai#access-panel) · [Requested URL](https://devin.ai/pricing) · SWE-1.7, coding-delegated, baseline, record `7e716da60646f690`. The host also has 30 successful operations across the cohort; these need not be the same URL.
- [redfin.com: 4 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=rate_limit&search=redfin.com#access-panel) · [Requested URL](https://www.redfin.com/blog/california-real-estate-commissions/) · SWE-1.7, residential-brokers, baseline, record `313eaa6ef8b7c812`. The host also has 27 successful operations across the cohort; these need not be the same URL.

### Unsupported content type — 93 (1.1%)

All 93 report text/markdown, and all are in Codex: 76 Astra, 17 Sol. Sentry contributes 30, Neon 14 and Prisma nine. This concentration suggests reader/representation compatibility rather than a general failure to fetch documentation.

**Website owner:** Respect requested representations, keep HTML available and separate cache variants. Do not assume an AI user agent supports Markdown.

**Harness / browsing-tool owner:** Support the received format or explicitly negotiate one the reader can parse. Record request and response headers.

[Inspect all matching observations](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=content_type#access-panel)

- [docs.sentry.io: 30 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=content_type&search=docs.sentry.io#access-panel) · [Requested URL](https://docs.sentry.io/product/issues/) · GPT-5.6 Sol, observability, baseline, record `3c15812891499bb6`. The host also has 17 successful operations across the cohort; these need not be the same URL.
- [neon.com: 14 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=content_type&search=neon.com#access-panel) · [Requested URL](https://neon.com/pricing) · GPT-6 Astra, managed-databases, baseline, record `6f60058dcbcfc200`. The host also has 36 successful operations across the cohort; these need not be the same URL.

### Invalid tool arguments — 35 (0.4%)

27 of 35 occur in Opus and eight in SWE. Twenty-nine lack a URL. The underlying audit found malformed search JSON and unsupported fetch fields such as offset and limit: these requests can fail before any network access.

**Website owner:** No website change indicated: invalid research-tool arguments can fail before a request is sent.

**Harness / browsing-tool owner:** Validate the tool schema and retry with supported arguments. Keep these errors separate from site availability.

[Inspect all matching observations](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=invalid_input#access-panel)

- [amplitude.com: 1 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=invalid_input&search=amplitude.com#access-panel) · [Requested URL](https://amplitude.com/pricing) · SWE-1.7, product-analytics, baseline, record `93322ade20c22bb4`. The host also has 71 successful operations across the cohort; these need not be the same URL.
- [daily.co: 1 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=invalid_input&search=daily.co#access-panel) · [Requested URL](https://www.daily.co/pricing/video-sdk/) · SWE-1.7, realtime-media, baseline, record `5f0db8cb84d3fb7c`. The host also has 50 successful operations across the cohort; these need not be the same URL.

### HTTP 401 — authentication required — 7 (0.1%)

All seven occur in SWE: six target Hugging Face and one api.openai.com. This is a small cluster of potentially gated resources and an API endpoint, not broad evidence that public documentation is inaccessible.

**Website owner:** Keep intended authentication; offer public descriptive pages and clear authorized-access instructions.

**Harness / browsing-tool owner:** Use public documentation as the research source. Do not treat credentialed API endpoints as ordinary pages.

[Inspect all matching observations](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=http_401#access-panel)

- [huggingface.co: 6 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=http_401&search=huggingface.co#access-panel) · [Requested URL](https://huggingface.co/briaai/BRIA-3.2) · SWE-1.7, model-text-image, baseline, record `0de6575e4de9df3e`. The host also has 644 successful operations across the cohort; these need not be the same URL.
- [api.openai.com: 1 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=http_401&search=api.openai.com#access-panel) · [Requested URL](https://api.openai.com/v1/models) · SWE-1.7, model-vision, comparative-selection, record `9c4471cc8316ab66`. The host also has 0 successful operations across the cohort; these need not be the same URL.

### Response too large — 6 (0.1%)

All six occur in Codex. Cohere reference pages account for three and Claude API documentation for one; two have no resolved host. The small sample points toward documentation payload limits, not a general site availability problem.

**Website owner:** Offer focused, bounded documentation pages and avoid unnecessarily large application payloads.

**Harness / browsing-tool owner:** Report the reader’s byte limit and support scoped retrieval where available.

[Inspect all matching observations](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=too_large#access-panel)

- [docs.cohere.com: 3 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=too_large&search=docs.cohere.com#access-panel) · [Requested URL](https://docs.cohere.com/reference/embed) · GPT-6 Astra, model-embeddings, tradeoff-led, record `c11e92d3bb8c8aa5`. The host also has 176 successful operations across the cohort; these need not be the same URL.
- [platform.claude.com: 1 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=too_large&search=platform.claude.com#access-panel) · [Requested URL](https://platform.claude.com/docs/en/api/messages) · GPT-6 Astra, model-chat, direct-fit, record `5fdc4e9b1f9f5e55`. The host also has 434 successful operations across the cohort; these need not be the same URL.

### HTTP 500 — server error — 4 (0.0%)

All four occur in SWE; three point to Technavio and one to an insurance site. These are repeated observations at two hosts, not four independent incidents. The origin versus intermediary responsible is not identified.

**Website owner:** Check application and gateway logs, dependencies and route health for the recorded time.

**Harness / browsing-tool owner:** Retain status and error details, distinguish intermediary errors and use conservative retries.

[Inspect all matching observations](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=http_500#access-panel)

- [statistics.technavio.org: 3 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=http_500&search=statistics.technavio.org#access-panel) · [Requested URL](https://statistics.technavio.org/it-research) · SWE-1.7, market-research, practitioner-choice, record `e85f408b90c2468c`. The host also has 0 successful operations across the cohort; these need not be the same URL.
- [itbusinessinsurance.com: 1 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=http_500&search=itbusinessinsurance.com#access-panel) · [Requested URL](https://itbusinessinsurance.com/) · SWE-1.7, insurance-business, baseline, record `5ba78ba8395dfc0e`. The host also has 0 successful operations across the cohort; these need not be the same URL.

### Access challenge in returned content — 4 (0.0%)

All four detections occur in SWE: three on an SF Chronicle article and one on PyPI. The reader got a response, but it asked for JavaScript or human verification. This illustrates why HTTP success and usable research content differ.

**Website owner:** Where automated public reading is intended, provide a supported static path or documented API.

**Harness / browsing-tool owner:** Detect challenge content rather than counting any returned body as a successful read. Respect access policy.

[Inspect all matching observations](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=challenge#access-panel)

- [sfchronicle.com: 3 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=challenge&search=sfchronicle.com#access-panel) · [Requested URL](https://www.sfchronicle.com/california/article/home-insurance-policy-19429777.php) · SWE-1.7, insurance-home, baseline, record `2b40ec7727a27221`. The host also has 0 successful operations across the cohort; these need not be the same URL.
- [pypi.org: 1 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=challenge&search=pypi.org#access-panel) · [Requested URL](https://pypi.org/project/instructor/) · SWE-1.7, ai-sdk, baseline, record `db0561c6548f25ea`. The host also has 4 successful operations across the cohort; these need not be the same URL.

### HTTP 402 — payment-related status — 4 (0.0%)

All four occur in SWE: three on docs.sweep.dev/pricing and one on Investopedia. Four status observations do not establish four paywalls; 402 has no standardized payment workflow and site implementations vary.

**Website owner:** Inspect the actual response and configuration; publish clear public pricing/access information where intended.

**Harness / browsing-tool owner:** Retain the response. Do not infer a standardized paywall workflow from status 402.

[Inspect all matching observations](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=http_402#access-panel)

- [docs.sweep.dev: 3 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=http_402&search=docs.sweep.dev#access-panel) · [Requested URL](https://docs.sweep.dev/pricing) · SWE-1.7, coding-delegated, baseline, record `a231e48eab99cf3f`. The host also has 0 successful operations across the cohort; these need not be the same URL.
- [investopedia.com: 1 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=http_402&search=investopedia.com#access-panel) · [Requested URL](https://www.investopedia.com/the-best-robo-advisors-8764849) · SWE-1.7, robo-advisers, comparative-selection, record `2113f6a12e60c74f`. The host also has 0 successful operations across the cohort; these need not be the same URL.

### HTTP 503 — unavailable service — 2 (0.0%)

Both observations occur in SWE at prompts-gpt.com, on the homepage and pricing page. This is one-host evidence with too little coverage to infer a broader trend.

**Website owner:** Check maintenance, origin health and upstream availability; provide retry guidance for temporary outages.

**Harness / browsing-tool owner:** Use bounded retry/backoff and retain the initial failure rather than replacing it with the retry outcome.

[Inspect all matching observations](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=http_503#access-panel)

- [prompts-gpt.com: 2 matching failures](https://aeo.latent.space/sources/?debug=1&view=access&access-scope=all&access-status=http_503&search=prompts-gpt.com#access-panel) · [Requested URL](https://prompts-gpt.com/pricing) · SWE-1.7, aeo, comparative-selection, record `71416fce22c6fe61`. The host also has 0 successful operations across the cohort; these need not be the same URL.

## Maintainer priorities and limits

Diagnose exact URLs against server logs, preserve stable routes, test HTML/Markdown negotiation and caching, retain intentional access controls, and verify fixes with bounded canaries. The live Sentry check described on the page is separate from the historical cohort.

Counts include retries; transports expose different detail. The 541 other errors are unresolved. The 192 unknown outcomes and 10 empty results are not counted as failures.

[HTTP negotiation and status semantics](https://www.rfc-editor.org/rfc/rfc9110.html) · [429 rate limiting](https://www.rfc-editor.org/rfc/rfc6585.html#section-4)