Evaluation sets — the machine is tested, never the person.

A Set of scenarios, on the Machine track only.

Family 0 is a Set on every track of every sector: the entry gathers other things rather than covering a machine, a seat or a kind of record of its own. On the Hands and Records tracks the things gathered are numbered skill sets. On the Machine track a Set may instead gather scenarios, and that is an Evaluation set. It exists on the Machine track only, because a machine is the one thing on this register that can be put through a scenario and observed.

A scenario is a situation a machine can be put in — a family of machine, an attachment, the ground, the material, an operating mode, the conditions around it — together with a stated method for observing what the machine does in it. It is set down and agreed by experienced people: the hands who know what the work looks like when it is done right, and the makers who know what the machine is for. They set it down the way any member of a Set is set down, and the register numbers what they set down.

Like every Set, an Evaluation set reads in order on its Card: Collected by, then Owned by Even Steven, then Licensed by Tacit Tessera. The Card says which kind of Set it is, and the tile's edition line says the same.

The machine is tested, never the person.

An Evaluation set tests a machine: a package running on a family of machine, in the scenarios the set holds. It does not test the person whose way of working the package is taught from, it does not test the person in the seat, and it does not test the people who set the scenarios down. No result of an Evaluation set is a fact about a person, and none is recorded against a person's number.

The rule that holds on every track holds here without exception: a record is never supplied to assess the person, to an insurer or to anyone else. There is no score, no pass mark, no grade, no star and no rank — not on a Card, not in the Book, not in any answer the Verify API gives — and no field anywhere compares one person with another. Protect carries the rule; Not a score says it in plain words.

Method only, never a result.

An Evaluation set's Card — the public side

Field	What it carries
Kind	Evaluation, on the Card and on the tile's edition line. Track M, family 0.
What it tests	The family of machine, the attachment and the work the scenarios are written for.
How	The method: the scenarios by name, the conditions each one states, and how what the machine does is observed and recorded. Never what any machine did.
Scope and Refuses	The context envelope: what the set covers and what it refuses, as one object.
Conditions	What each contributor has put on what is made of their work, carried from the member.
Collected by · Owned by · Licensed by	The collector's mark; Owned by Even Steven; Licensed by Tacit Tessera.
Status	Specified — written down and numbered, not issued.

What the register records about a test is who tested, on what date, and against which stated method. The result itself is the licensee's, about its machine: the register holds none and publishes none. Never a result. Never a figure. Never a person.

What a set covers, what it refuses, and the claim it does not make.

Every open question on this site is written the same way: the question, the fact that it is open, and what is built.

The question. Can a machine's learned behaviour be validated for a field of work — declared fit for a family of machine, a ground, a material, a task — by running it through a set of scenarios?

It is open. There is no field-wide standard to validate a machine's learned behaviour against. A test says what a machine did in the scenarios it is given; it does not say what a machine will do in a scenario nobody wrote. So validated for is not a claim this register makes, nor is any word of that shape, and no Card carries one.

What is built. A set states what it covers and what it refuses, as one object — machine families, attachments, environment, materials, operating modes, jurisdiction, stated assumptions, known exclusions. A machine is tested inside that envelope, and the result, to the licensee, is about what happened inside it. Outside the envelope the set says nothing, and says so. The register's one public status word is Specified, and it stays that on an Evaluation set as on every other entry.

An execution licence, to whoever has a machine to test.

Three kinds of licensee would license an Evaluation set. A machine maker, to run a package through the scenarios on its own family of machine before a release is pinned. A kit company, whose kit takes the seat of a machine already in the yard and which needs to know how that seat behaves before the kit ships. And an AI that serves an industry — a platform, an agent builder or a command centre — that runs a Machine skill set and needs the same answer about the same machine.

The licence kind is execution: the right to run the set, with a term, a field of use and a territory, counted on every run. An Evaluation set is run against a machine; nothing is built from it, and it is never loose material. The count is the economic record, and the rate lives on Even Steven's covenant and never on a page. The licence, kind by kind, is on License; the count on Meter.

Each scenario keeps its own member's ending, as on any Set. When a member is withdrawn, the ruled sentence holds: It stops for new work at once; a job already underway finishes. A test already running runs out; nothing new starts from the day the withdrawal arrives. A contributor's conditions travel with a scenario as with any member: the field of use, the territory, no use for surveillance, no use where a machine runs unattended, and a named use or house kept away from it.

Found on the Exchange, offered as a Collection, shown as what it is.

An Evaluation set is found on the Exchange by the public side of its Card — number, name, status, tile, scope statement — and by nothing else. There is no list, browse, bulk or export endpoint. Offered under one paper it is a Collection, shown as the kind it is, beside the Mosaic and the Composite. What the Exchange shows of an Evaluation set is what it tests and how. No result travels with it, on the Exchange or anywhere.

A check at tacittessera.com/verify answers four things about its number and nothing else, and the Verify API refuses any evaluation result, to anyone.

Open, with nothing filed.

No Evaluation set exists. No scenario has been set down, no machine has been tested against one, no result has been produced and none has gone to anyone. The kind is specified before the first is written, so that the first is written to it, and this page is that specification.

A number filed here would be read like this one: C M 0 001 GN · 1, in machine form CM0001GN14 — sector C for Construction, track M for Machine, family 0 for a Set, the general trade, the edition, and a check character the register computes and nobody types. That number is an illustration, not an entry in the Book.