vitna.ai/docs/concepts/evidence Machine view
Raw .md llms.txt
> Source: https://vitna.ai/docs/concepts/evidence
> Raw Markdown: https://vitna.ai/docs/concepts/evidence.md
> Corpus index: https://vitna.ai/llms.txt

# Evidence

The archived document a figure was read from, addressed by the sha256 of its
bytes.

Every priced cell on the tape carries an `evidenceSha256`. That hash is the
link between a number on a page and a document that existed.

## What is public and what is gated

Deliberately split:

**Public**, for any hash, with no key:

- the URL that was fetched
- the time it was fetched
- the HTTP status and response headers
- the user agent, the adapter, the parser version, the byte count
- what robots.txt said
- every subsequent read that found the same bytes

**Gated** behind a bearer token: the archived copy of the document itself.

```bash
curl https://vitna.ai/api/v1/evidence/<sha256>
```

## Why that split

The public half is what a verification actually needs. Given the URL and the
fetch time, you can fetch the page yourself and compare hashes. Vitna does not
have to be trusted for that check to work, which is the entire design goal.

The archived copy is a third party's page. Republishing it wholesale raises a
copyright question that this record does not need to answer in order to be
checkable, so it is held and not served openly.

## What a hash proves

That the bytes Vitna read hash to this value, and that a figure was parsed out
of them at a stated place by a stated parser version.

It does not prove the provider published the price. Nobody signs their own
pricing page. What ties a figure to a provider is the archived document plus
the URL and time it was fetched from, and those record what Vitna fetched.

That distinction is stated in the API responses themselves rather than only
here, because it is the claim most likely to be overstated by someone quoting
the record.

## Where the archive lives

On disk, at `VITNA_EVIDENCE_DIR`, defaulting to `/srv/vitna/evidence`. Each
document has a sidecar holding the fetch metadata above.

Documents are stored gzipped. The hash is of the document as fetched, before
compression, and the byte count is its uncompressed length, so the body the
endpoint returns hashes to the key you asked for. Documents archived before
2026-10-14 were stored uncompressed, and both forms are read.

The read that first archived a document is on the sidecar. Every fetch that
has since found the same bytes is in the database, and both come back together
from the evidence endpoint.
Human Machine Shift+M