Template reuse

How much of this project's copy is not its own

Paste a Solana token mint address and Evidesa reads the links its metadata declares, requests at most three of the project's own pages, and measures how much of their wording is shared with the reference entries configured in this deployment.

Or name two tokens and compare them directly. That is the usable form of this surface today, and the one that establishes something on the first try.

The reference list ships empty, and there is no plan to fill it from a scrape. An entry in it is a factual claim about a named project, recorded on a date, for a reason a reader could check — and a similarity measurement is not that reason. With no entries configured, a run against the list alone reports that nothing was compared. That is the honest state of this surface rather than a defect, and it is not a clean bill of health for any token put through it. Naming the second token yourself is a different act: you choose both sides, Evidesa keeps no record of either, and the answer is about the two addresses you asked about rather than about a list somebody else drew up.

Running this fetches up to three of the pages this token's metadata points at. Each request tells that page's host that Evidesa asked for it.

Leave this blank to compare against the reference entries configured here, of which there are normally none. Name a second token and both are read and measured against each other. Doing so fetches that token's pages too, and tells those hosts that Evidesa asked for them. A result says the two publish similar wording — never that they share an author, an owner, or a purpose.

PDFs are read, or refused. A whitepaper served as a PDF is extracted, including the embedded fonts most of them use — measured against seven published whitepapers, all seven read correctly. A PDF stores glyph codes rather than words, so where a document's encoding cannot be mapped back to characters, extraction would produce text of the right length with the wrong letters. That output would still yield a similarity number, so it is refused instead and the document is reported as unread. Encrypted PDFs and pages that are scanned images refuse the same way. An unread whitepaper is not a whitepaper that says nothing.

What a similarity measure means. Shared copy is evidence of shared copy. Nothing more is established by it: not shared ownership, not shared operators, not coordination, not intent, and no wrongdoing by anyone. It does not establish who wrote either document, or which of them came first.

Shared copy is ordinary, not remarkable. Templated launches are the normal case. One launchpad theme, one site builder, one landing-page kit, one paragraph both sites took from a third — any of these produces near-identical pages for people who have never met, and the measurement cannot tell that apart from deliberate reuse. Two projects using the same template may have no connection at all.

The numbers are not calibrated. A measure of 0.80 is not eighty per cent of anything, and the distance between two of these values carries no fixed meaning. There is a reporting threshold, chosen for what it keeps out rather than what it lets in, and a measure below it is not a finding in either direction.

Fetching a page is a visible act. Requesting a project's website tells that website's host that Evidesa asked for the page: the request arrives from this deployment's network address, at whatever time it was made, into whatever logs that host keeps. At most three pages are read per run, one after another, once — no caching, no repeat crawling, and a per-token cooldown so the same pages are not requested again for a reader who arrives a minute later. Social platforms and code hosts are not requested at all, because an anonymous request to them returns the platform's own markup, which every project on it shares.

What those pages said is not reproduced here. Page text is measured on the server and discarded. It is written by whoever runs that site, and printing it inside this page would let a third party make claims in Evidesa's voice — the same reason the transaction pre-check reports a memo by its length instead of showing it. What comes back is the measurement, which addresses were requested, and which were not.

What is not examined. Logo and image similarity is not evaluated in this version, and neither is the text of any social profile bio. A roadmap presented as an image, a slide deck, or a document format this version does not read is invisible to it. A page that did not respond, redirected somewhere this version declines to follow, or returned an encoding it will not decode is reported as unread. Unavailable evidence is not evidence of low risk.

This produces no risk score. No measure on this page contributes to any of the five risk categories, to a token's overall risk figure, or to its Evidence Dossier. It is a separate, non-risk observation about text, and it is kept separate so that a number which cannot carry a verdict is never read as one.

On naming anyone. A reference entry describes a record — who established it and how — never the people behind a project. Nothing on this page accuses any named party of anything, and a shared five-word window could not support such a claim if it did.

Evidesa is not financial advice and does not certify safety.