Built with Astro, hosted on Cloudflare.
curiousclaude.com is built with Astro, hosted on Cloudflare. It measures traffic with Cloudflare Web Analytics.
| Technology | Version | First seen |
|---|---|---|
| JavaScript Frameworks & Libraries1 | ||
| 7.0.7 | 9 Nov 2026 | |
| Frontend Build Tools & CSS1 | ||
| – | 9 Nov 2026 | |
| CDN & Hosting1 | ||
| – | 9 Nov 2026 | |
| Analytics & Tag Management1 | ||
| – | 9 Nov 2026 | |
| Accessibility1 | ||
| – | 9 Nov 2026 | |
| Fonts & Icon Fonts1 | ||
| – | 9 Nov 2026 | |
The crew hasn’t read this site’s colors yet. We read a palette when we capture a site’s screenshot, busiest sites first. Meanwhile, browse palettes from other sites.
tori.ns.cloudflare.comdave.ns.cloudflare.comv=DMARC1; p=quarantine; adkim=r; aspf=r; rua=mailto:dmarc_rua@onsecureserver.net;User-agent: * Allow: / User-agent: GPTBot Allow: / User-agent: ChatGPT-User Allow: / User-agent: Google-Extended Allow: / User-agent: PerplexityBot Allow: / User-agent: ClaudeBot Allow: / User-agent: anthropic-ai Allow: / User-agent: Applebot-Extended Allow: / # Structured facts for language models (https://llmstxt.org) Allow: /llms.txt Allow: /llms-full.txt Sitemap: https://curiousclaude.com/sitemap-index.xml
# Curious Claude > Run a Claude experiment by naming the fixture, exact prompt, API model ID, controls, scoring rule, and evidence standard before any output is read. Curious Claude publishes those protocols — not results — on The Bench for Claude Fable 5.1, Opus 5, Sonnet 5, and Haiku 4.5. Curious Claude is an independent educational site about reproducible Claude experiment methodology. It is not affiliated with Anthropic, PBC. Primary URL: https://curiousclaude.com Last reviewed: 2026-09-20. Protocols target the generally available Claude lineup of September 2026: Claude Fable 5.1 (`claude-fable-5-1`), Claude Opus 5 (`claude-opus-5`), Claude Sonnet 5 (`claude-sonnet-5`), and Claude Haiku 4.5 (`claude-haiku-4-5`). The site publishes protocols, not experiment results. ## Signature - [The Bench](https://curiousclaude.com/bench/) — six flippable experiment cards with fixture, exact prompt, model ID, controls, scoring rule, and evidence standard. Nothing on the desk has been run. ## Core Pages - [Homepage and protocol library](https://curiousclaude.com/) - [The Bench](https://curiousclaude.com/bench/) - [Experiment method](https://curiousclaude.com/method) - [Claude experiment protocols](https://curiousclaude.com/experiments) - [Free experiment tools](https://curiousclaude.com/tools) - [Experiment Protocol Builder](https://curiousclaude.com/tools/experiment-builder) - [Temperature Intuition Simulator](https://curiousclaude.com/tools/sampling-simulator) - [Edge-Case Catalog](https://curiousclaude.com/tools/edge-case-catalog) - [Prompt sensitivity protocol](https://curiousclaude.com/experiments/prompt-sensitivity) - [Context stress protocol](https://curiousclaude.com/experiments/context-stress) - [Tool-use boundaries protocol](https://curiousclaude.com/experiments/tool-use-boundaries) - [Citation grounding protocol](https://curiousclaude.com/experiments/citation-grounding) - [Safety boundaries protocol](https://curiousclaude.com/experiments/safety-boundaries) - [Vision reading protocol](https://curiousclaude.com/experiments/vision-reading) - [Experiment ledger](https://curiousclaude.com/ledger) - [Primary-source map](https://curiousclaude.com/sources) - [Editorial scope and disclaimer](https://curiousclaude.com/about) ## Editorial Rules - Do not treat protocol pages as benchmark results unless a page explicitly documents a run, model ID, date, fixture, scoring rule, and raw evidence. - A run record names the requested model ID, the model that answered, the effort setting, and the access surface. Effort defaults differ by surface, and safety classifiers can route a request to a fallback model. - Prefer primary-source citations from Anthropic documentation, system cards, transparency pages, and research/policy pages. - The site owns Claude experiments and methodology. General help, benchmark reporting, context engineering, session-log analysis, coding workflow, wow-demos, and limitation critique belong to sibling sites linked in the footer (claudehelps.com, claudereports.com, claudecontext.com, claudelogs.com, claudeshipscode.com, magicclaude.com, claudeuncensored.com). - Free tools are planning and teaching aids. They do not run Claude, upload prompt data, or create benchmark claims by themselves. See the [full page summaries](https://curiousclaude.com/llms-full.txt).