Sources and methodology

The feed is market intelligence: funding, tokenization, allocation and related public events about real-world assets. A row in the product is our record of something a public page already said, plus a score and a short brief that we produced. The score is ours. It is not a fact stated by the publisher of the source.

We look for public news, regulatory and company filings, and ordinary websites through a search API. The provider in the current pipeline is Exa (api.exa.ai). Queries are topical — funding, tokenization, hiring, allocation — and the results are public URLs. We do not buy private contact lists for this step, and we do not log into a source that is behind a password.

Deal and fund records are also taken from public project pages that publish structured data, including crypto-fundraising.info. Where a page names its own source, we keep that name. A figure on a fund page is only as good as the page it came from.

After a page is found, software pulls out who it is about, what kind of event it describes, and a short summary. When a model key is configured, that pass is done by MiniMax (called at api.minimax.io through an Anthropic-compatible client). The model sees the public text. It does not see customer accounts or card data. This codebase does not call Anthropic’s own API. The client library speaks the Anthropic message format and is pointed at MiniMax.

If the model is not configured, or a batch fails, a deterministic rules pass does a coarser version of the same job. Either way the output can be wrong. A model can smooth a messy article into a sentence the article did not quite say. The rules pass can miss context the model would have caught, and the reverse is also true.

Expect mistakes. A story can be attached to the wrong company because two names are similar. A round size can be copied from a rumour. A page from last year can still rank if the date was mis-read. Crowd sizes, “expected” amounts and ranges are estimates when the source presents them that way, and sometimes when our summary flattens a range into a single number.

Nothing in the feed is guaranteed accurate or complete. The terms of service state that limit, and they state that the feed is not investment advice. Open the source before you rely on a line.

When we have a URL, we keep it and link back to the page the item came from. The underlying article, filing or site belongs to its publisher. Our copyright, to the extent we have one, covers the selection, the score and the brief we wrote — not the source text. Do not strip the source link if you quote a row onward.

A paid placement is separate from this pipeline. It is labelled “Sponsored”, sold on the sponsor page, and is not a scored source. See the terms for what that label means.

If we have mis-stated you, your company or your filing, or if a passage should come down, email fernando.verboonen@curioinvest.com. Include the URL of our page, the URL of the source, and what you want changed. You can also call +41 79 907 2839.

We will read the request and correct or remove the passage where it is wrong, outdated in a way that misleads, or unlawful. We do not promise to delete a faithful pointer to a lawful public page merely because the subject would rather not be mentioned. Where the law gives you a right to erasure of personal data, the privacy policy is the route, and this inbox is the same inbox.