Intake open · assay in about a minute

What is your data worth to an AI lab?

Upload a sample. Get it graded on the seven things a lab checks before it pays — and listed where their researchers search.

Assay record — example LOT 26-0412
Rights & provenance88
Scarcity91
Process signal76
Fidelity83
Training-readiness54
Coverage69
Refresh40
▲ the level the trade requires s2lB26
01 / INTAKE

Upload a sample. Get the assay.

Twenty rows of the real thing is enough. Nothing is listed until you say so.

01Sample
02Provenance
03You

Drop in a slice of the real thing

No account, no email. We read the first 6,000 characters.

or drop it here
CSV · JSONL · JSON · TXT · MD · TSV
Include the messy fields — we score what a buyer receives.
Unclaimed uploads are deleted after 72 hours.

Now the part a sample can't show

Rights, coverage and refresh decide 42% of the score and none of them are visible in the data.

Where should the assay go?

The full report takes about a minute. We'll email it.

or use email
You keep every right you walked in with.
A listing shows scores, category and volume — never the data.
Buyers come to you. We don't broker and don't take the contract.

Seven dimensions, against the buyer bar
Hard gates
Three things that would move this
    Indicative band

    How a lab would read it

    02 / DEMAND

    What labs are hunting.

    The web kept the finished output and threw away the process. The process is what's scarce.

    CategoryDemand90-day

    Seeded from public licensing activity through mid-2026. Buyer briefs replace these numbers as they land.

    03 / RUBRIC

    What the assay scores.

    Published on purpose. Read it, fix your data, then submit.

    24%Rights & provenanceCan you prove where it came from and licence it for training?
    20%ScarcityCould a buyer get substantially this for free?
    14%Process signalReasoning, tool calls and failed attempts — or only the polished result?
    14%FidelityAccuracy, label noise, and whether claims are checkable.
    10%Training-readinessSchema, encoding, duplicates, format. Cleaning time comes off your number.
    10%CoverageSpread across the space you claim to cover.
    8%RefreshFrozen archive, periodic drop, or live feed. Feeds price as subscriptions.

    Five ways to fail regardless of score

    Personal data — identifiable people, no lawful basis.
    No chain of title — you can't document how you got it.
    Somebody else's content — scraped or rights-reserved, repackaged.
    Already in the corpus — duplicated from what buyers already hold.
    Terms that forbid it — your own agreements block sublicensing.
    Fail one? — usually paperwork. The assay says which, then re-run.
    Do you take ownership of my data?

    No. The sample produces your assay, and nothing is listed until you press the button. A listing carries the description and the scores, never the data. Buyers contact you and you negotiate directly.

    How much should I ask for?

    Anchor on what it replaces. If a lab would otherwise pay experts to produce the equivalent, price against that — $30 an hour for reviewers up to $200 for specialist clinicians. Exclusivity roughly doubles the conversation.

    What if my dataset is small?

    Small and unique beats large and ordinary. A few thousand genuinely non-public records capturing expert judgement will get a meeting; ten million scraped listings won't.

    04 / THE INDEX

    Describe the data you need. In a sentence.

    Every listing has been through the same assay, so scores compare across sellers.

    Sign in with your lab email

    The index is closed. Sellers list what they hold and what it's worth — reading that takes an account we can attribute to a buyer.

    or