Seller intent

AI data licensing glossary

Last updated November 1, 2026/Seller intent

This AI data licensing glossary fixes a small set of words the rest of the site uses, so a technical reader and an owner are not talking past each other. A trajectory is a path. Tool use is one kind of step on a path. An RL environment is a task with a check. An evaluation set is a group of past cases held aside. A license is a permission. A sale of the data is a different transaction. None of these words means the company has already agreed, and none of them means a use is lawful. The definitions below point at the page that does the real explaining.

Three pillars hold the map. Commercial permission is what AI data licensing means for a company. How operating records can teach an agent to act is how business data trains AI agents to do real work. What a checkable task still has to be, beyond a folder of files, is what reinforcement learning environments need from real work. If you only remember three links, remember those.

Why the words have to stay narrow

Teams waste months when log, dataset, and environment are used as synonyms. A buyer asks for an environment and receives a zip of reports. An owner hears license and signs language that assigns the archive. Narrow words are how you notice the mismatch before files move. The value of a glossary is not elegance. It is the chance to stop.

Stopping is also a rights decision. A term can be defined perfectly and the underlying records can still be forbidden by a contract or a notice. Definitions do not clear permission. Why legality is not a yes for every company is the page that refuses a blanket answer. This glossary is not legal advice.

The terms, and where each one is unpacked

Trajectory

A trajectory is an ordered path: a situation the actor could see, an action, and what came back, repeated until an ending or a stall. It is evidence of work. It is not a task you can reset. The full account is what an agent trajectory records from real work. The business logs that sometimes contain a path are how workflow and process records can be licensed.

Tool use

Tool use is the step where software was asked to do something and returned a result the person then accepted or refused. It is not a list of subscriptions, and it is not a password. See why records of software tool use matter to agents. Providers who live in other companies' consoles should also read what a managed service provider already generates that can train AI, because credentials are out of scope there.

RL environment

An RL environment, in the sense this site uses, is a named task: a start, allowed moves, and a check a second person could audit. Company history can inform that task. History is not the task. The pillar states the requirements. How a business workflow becomes a checkable training environment is the translation of one process into those three names, including when to abandon the translation.

Evaluation set

An evaluation set is past cases with known outcomes, withheld so an agent can be tested rather than taught. If the case was already in the teaching slice, it is not a clean test. The write-up is how past business outcomes become agent test cases.

Expert workflow

Expert workflow data is a procedure, a case, and a recorded decision to follow, bend, or stop, with a reason. A résumé is not that record. See how expert judgment and decision records can train AI. Written procedures as a shelf are how internal documents and playbooks can be licensed.

License and sale

A license is permission for a defined use. The company keeps the records. A sale of the data moves control of the dataset. This marketplace runs licenses, not purchases of your files. The comparison is how a license differs from selling the data. Scope, term, exclusivity, deletion, and audit are the topics a draft should make readable, listed on what a data licensing agreement should cover.

Profile

A Data Partnership Profile is a description of shelves, years, and limits. Buyers who are shown anything early see the profile, not the raw files. It is not a delivery and not an offer. How a company gets to that description without an upload is how a company gets records ready for a buyer.

Misreadings that cause bad conversations

• Calling every export an environment. If nobody has named a start, the moves, and a check, the word is false.
• Calling a license a sale, or a sale a license, because the search box used both. The grant in the document is what counts, not the headline.
• Treating an evaluation set and a training slice as interchangeable. One tests. One teaches. Mixing them hides failure.
• Assuming a defined term is a cleared right. You can define a trajectory and still be forbidden to offer the log it came from.
• Using industry labels as if they were permission. A plant, a jobsite, a firm, and a software company each have different walls. The industry pages exist so those walls are not collapsed into one slogan.

How to use the glossary

Big Data/Little Guy, operated by Unlockt, LLC, keeps these words stable so a profile does not silently change meaning between conversations. If you want a commercial reading of the business you run, use the free data value check. The check does not move files and it does not assign any of the terms above to your archive.

Bring the word you were sold. Look it up here. Then open the linked page and see whether the object you actually have matches. If it does not, say so. A mismatch caught in a glossary is cheaper than a mismatch caught after a draft. If counsel needs to read a contract, this page will not replace that reading.