The agent successfully assembled a comprehensive explanation of DonationWatch by combining sparse website content with a machine-readable API document and data index. The site is moderately agent-ready: it publishes technical specs via llms.txt and exposes a JSON data index, but core marketing and pricing information remain fragmented, incomplete, or deliberately gated behind contact forms, forcing the agent to rely on inference and prior knowledge to build a coherent answer.

• ›Step [9] (llms.txt) was the single most content-bearing fetch: it explicitly stated what DonationWatch does ('neutral, data-driven political party donation tracker'), who it's for (researchers, AI agents), and its commercial model (free public + experimental Enterprise API). This document was not discoverable through the website UI and had to be guessed.
• ›Step [13] (data/index.json) revealed the multi-country scope and data structure (Germany, Austria, EU, etc. with date ranges), answering the 'what countries' question that the website itself did not clearly list.
• ›Steps [2] and [20] (about and other-countries pages) were fetched but returned truncated HTML; the agent could not extract their actual text content, indicating the site uses client-side rendering that was not resolved in the fetch. The agent worked around this by leaning on llms.txt and the JSON index.
• ›Enterprise pricing is deliberately hidden: step [3] (/pricing) returned truncated HTML with no pricing data visible. The agent correctly inferred from llms.txt that enterprise pricing exists but is not public and requires contact.
• ›The site does not provide a competitor comparison or historical depth documentation; the agent inferred differentiation (multi-country, open-source, AI-optimized) from combining llms.txt, data index structure, and prior knowledge of OpenSecrets/FollowTheMoney.
• ›No web search results or GitHub content directly answered the task—searches returned homepage URLs or unrelated tools. The agent's success relied entirely on guessing the existence of llms.txt and the JSON data endpoint, both of which are developer-facing and not linked from the public UI.