- File
- —
- Size
- —
- Detected type
- —
- SHA-256
- —
- Chunk Merkle root (1 MiB leaves)
- —
- Entropy
- —
- Image
- —
- Metadata check
- —
- Not-before anchor
- —
DM—0000 · Frontispiece
Every file has a maker. We keep the receipt.
Datamint is an archive for human data that was given, not scraped. Contributors fingerprint their work in the browser, pick the terms it may be used under, and are paid each time a licence is sold.
Browse the catalogueFingerprint a file
Plates drawn in your browser from each catalogue entry's seed. No stock imagery.
i.
The catalogue
Twenty-four sample entries showing how a consented dataset is described: what was collected, from whom, under which licence, and what a buyer pays. These entries are illustrative samples, not datasets for sale.
ii.
Fingerprint a file
Drop any file. It is read and hashed on this device with the browser's own SHA-256; nothing is uploaded. You get a provenance and licence card, anchored to the latest finalised Solana block so it can't claim to be older than it is.
No file to hand? (a small CSV made in your browser).
Check a file against a card
Load a Datamint card, then the file it describes. We recompute the hash and check that the card itself hasn't been edited.
No card loaded.
iii.
Licence matrix
Six licence templates, eight questions buyers and contributors actually ask. Pick a use along the top to see which templates allow it, or answer the four questions below and the matrix narrows itself. Template terms are drafts for illustration, not legal advice.
Contributor's view: what should buyers be allowed to do?
All six templates match.
| Template |
|---|
allowedconditional (hover for terms)not allowedPrice multiplier is a model input used by the payout calculator.
iv.
Contributor payouts (a model)
How much a steady contributor might earn, shown in dollars and in SOL at the live price. Every input is visible and every rate is an assumption you can change. This is a model, not an offer or a forecast.
v.
From a file to a licence, in four acts
Capture
A contributor makes something in the course of their life or work: photographs of a repair, a recorded reading, a day of motion data. Nothing is collected passively or in the background.
Fingerprint
The file is hashed on their device. The hash, not the file, is the identity: the same bytes always give the same fingerprint, and one changed pixel gives a completely different one.
Describe & consent
The contributor chooses a licence template, confirms consent from anyone who appears in the work, and flags any personal data they removed. The card records all of it.
License & pay
Buyers license a dataset, not a person. Each sale is split by the share the contributor saw up front, and revocation, where chosen, stops future training runs from using the work.
“Provenance is not a promise. It is a hash and a block number you can check without asking us.”
Consent is a file, not a checkbox
Every entry that shows a person needs a consent record linked to the same fingerprint. No record, no listing.
Scrub before you ship
The fingerprinter flags EXIF blocks and location tags on photographs, because the most common leak in contributed data is the metadata nobody looked at.
Prices in daylight
The share a contributor receives is shown before they upload, and the calculator above uses the same formula, in the open.
Small, legible licences
Six templates instead of bespoke contracts, so that contributors can compare them in one table and buyers know what they bought.
vi.
Questions
Is my file uploaded when I fingerprint it?
No. The file is read into this tab's memory and hashed with crypto.subtle.digest('SHA-256'). The only network calls the fingerprinter makes are read-only Solana RPC requests for the latest block, which carry no information about your file.
What does the Solana anchor prove?
That the card was produced after a specific finalised block, because it quotes that block's hash, which nobody could know in advance. It does not prove the card existed at any later time: the card is not written on-chain. To prove that, a contributor would publish the card's digest in a transaction of their own.
Why a Merkle root as well as a SHA-256?
The file is split into 1 MiB pieces and each is hashed; the piece hashes are paired and hashed again up to a single root. A buyer can then check that one piece belongs to the file without downloading the whole thing. When a level has an odd number of hashes, the last one is paired with itself.
Are the catalogue entries real datasets?
No. They are illustrative samples written to show what a well-described, consented dataset looks like. Their plates are generated in your browser, and their prices are model outputs.
Where does the SOL price come from?
Jupiter's public price API, with Hyperliquid's mid price as a fallback. The 24-hour line in the masthead uses Hyperliquid hourly candles. Prices refresh every minute.
Is the payout calculator a promise?
No. It is a transparent model: accepted items accumulate in a library, each licence sold pays a per-item rate times the licence multiplier, and the contributor receives their share. Real demand varies widely, and many datasets never sell.
Can I take my work back?
Under templates marked revocable, yes, for future use: models already trained are not un-trained, but new training runs must exclude the work. The licence matrix shows which templates allow it.
Does Datamint hold tokens or funds?
Not on this site. Nothing here mints, trades, custodies or signs anything. It is a working catalogue design and a set of tools.