Every record in the catalogue is built from the same parts. That list of parts is the schema. It is what lets one record be compared with another, and what someone else would need to know to reuse the catalogue or to check our work.
What every record states
- What it does. One or two sentences on the job the tool does.
- Where it stops. The limit the tool keeps to, or the fact that it has none.
- How easily you can stop using it. What leaving costs, and whether you can repair the tool yourself.
- A position. A number from 0 to 100 on the threshold scale.
- A review status. Provisional, Reviewed, or Revised, with the date the record was last checked and a dated history of every change to its position.
- A catalogue number. Issued once, when the record is published, and never reused.
- A domain. The area of life the tool serves, such as Mobility or Learning.
- Systems. Any larger system the tool is part of, such as Schooling or Medicine.
These are stored as separate fields, so records can be listed, sorted, and exported. Two more parts are written in the body of each record: the six balances, each with a score and a sentence of evidence, and the sources. A tool that does not yet do its job still has its six balances recorded, but they do not move its position: below 30, the position is a judgment about whether it works, and the record says why. The position is worked out from the balances, by the rules in How we evaluate tools.
It is provisional
The schema is at version 0.1. The fields can still change, and some have. “Where it stops” was renamed and then renamed back. The rule for working out a position changed on October 6, 2026, so that a tool whose six balances all hold sits at the centre of the convivial range. When a record’s position moves, its review history says when and why.
What 1.0 will mean
Schema 1.0 is the point where the fields are frozen. After that, changing one is a breaking change, to be announced and explained. We will call it 1.0 when the fields have been tested against enough real records to trust them.
Open questions
- How should the evidence for a position be cited, so that each score points to its source?
- Should a record state the tool’s licence, who maintains it, and how repairable it is, as separate fields?
- Is a number from 0 to 100 the right model, or would three zones with notes be more honest?
- Do the watershed marks at 30 and 70 belong to the whole catalogue, to each domain, or to each tool?
- Should the left half of the convivial range carry meaning? Today a tool whose balances all hold sits at 50, and positions from 30 to 50 are unused. One proposal is a score for how reliably the tool does its job. That needs working out first: reliable for whom, and on what evidence?
- Should the balances and the sources become fields too, so they can be compared across records?
- Can the scale show who bears the cost? A society of many overlapping, local literacies may bear the costs of a tool differently from one administered in a single standard language. A position records how far a tool has gone, not where the harm falls or on whom. The question comes from the record for Literacy.
A proposal for 1.0: reliability as its own field
The scale is one line, and it measures one thing: how far a tool is from the second watershed. It cannot also show how well the tool works. A tool can be unreliable and have balances tipping at the same time, and adding one while taking away the other would put it at the centre, which is the wrong picture.
So the proposal is to record reliability separately, at three levels:
- Works for most people who try it.
- Works for some. It needs skill, setup, or local supply that many people lack.
- Works rarely. It cannot yet do what it claims.
The left half of the convivial range would be used only by tools with no balance past its limit. “Works for some” would place a tool at about 40. “Works rarely” would place it below 30, where carbon capture and medicine before 1913 already sit. A tool that is unreliable and has balances tipping would keep the position its balances give it, and show its reliability as a note.
Tools likely to land between 30 and 50 are of three kinds: ones that need skill, such as running your own server or a mesh network; ones that depend on local supply, such as a tool library, a repair café, or public transit in a small town; and young tools still proving themselves. Two existing records test the idea. Large language models are reliably fluent but not reliably right, and are already past the second watershed, with balances lost. The fediverse works well for people who get past choosing a server, and less well for those who give up there.
Nothing is decided. The proposal adds a second judgement to every record, and three questions come first: reliable at what, for whom, and on what evidence?
The code that defines the fields is free software. See the Workshop.