DAX Foundation

History

How this is paid for, and what we will measure

Part 5 of 5 · What has run

  1. The first run: a first page in eight minutes, and what it got wrong
  2. Fourteen small experiments, and why small first
  3. The stranger test
  4. How a showcase is made
  5. How this is paid for, and what we will measure

Meta DAX is free. It is built on paid compute. Those two sentences are how this is paid for, and the rest of this piece is what sits behind them: where support goes, what the Foundation does and does not take, what a fiscal sponsor would change, and then what we will measure. Nothing has been measured with a person yet, and everything after that depends on it.

Where support goes

There are two doors. Ko-fi takes a one-off coffee or a monthly one, with no GitHub account needed, and it goes to Jason Jeyanandan personally. GitHub Sponsors takes support monthly or once. The Foundation takes none of it. Open the Sponsors page and it says what the money is for, which is compute, and that when the compute runs out the building stops. We give no amounts here. Obsidian Delta, the company, sells done-for-you work. That is separate: it is the company’s, not the Foundation’s.

What a fiscal sponsor would change

The Foundation is not incorporated, and that matters to anyone who funds through a grant programme. The Saskatoon Community Foundation’s grant FAQ says that “most Saskatoon Community Foundation grants programs are open only to ‘qualified donees’ as defined by the Income Tax Act of Canada.” A programme drawn that way, as things stand, leaves out the Foundation. The same page names the usual route round: “In some cases, it may be possible to partner with a qualified donee which can act as a fiscal sponsor for the program to receive the funds” (Saskatoon Community Foundation, Grant FAQs).

So a fiscal sponsor changes who receives the money: a qualified donee does, on behalf of the work. It also adds a duty. The Canada Revenue Agency’s guidance on using an intermediary says that when resources are transferred to one, the party transferring them must “direct and control the use of those resources”. That is general guidance, not a description of any arrangement of ours, but it is why “fiscal sponsor” reads as a partnership with duties on both sides, not a mailbox. This describes how the arrangement works. It is not a plan, and nothing here says a sponsor exists.

What we will measure

Open Human delta value: the measure at “How it will be measured” and the section says that “Nothing in this section has been measured yet” and that “All of it is designed, not run”. Human delta value is what a person brings, alone and together with their companion. The measure names three records. The first is the human delta value record: a place in the session record for the person’s account, written as they gave it. It is designed, not built, and the measure says it “MUST exist before any experiment with a person runs.” The second is an override record from the planned nurse’s walkthrough, where a guide’s overrides of the system are recorded as theirs. The third is the guide’s own account at the end: what the system took off them and what they then spent their attention on, never inferred from logs and never reduced to minutes. A record like that has to live where it cannot be quietly edited. The Constraint Protocol is meant to be that place, and Meta DAX is meant to run on it; its protocol fields are reserved and empty today. A record, not a wall is the argument for keeping one.

None of the three tests the unit claim alone. The claim is that where a person brings what their companion lacks on a task, the two together reach a higher level than either alone. Thirty to One is the picture behind it, thirty seats behind one learner, and the one seat nothing else fills. The kill line is one measure: the level the unit reaches on the task, “scored blind by a separate grader”, compared with the levels the person and the companion each reach alone. The claim is false if, on tasks where the person alone scored above the companion alone, the unit does not score above both. It is also false if, in matched sessions, the unit reaches the same level whether the person brings their best or brings less. We measure level and not gain, because gain favours whoever brings less. No experiment has yet run the person alone, the companion alone and the unit on one task. The result will be published either way.

The first measurement is the first real families and the first ten scenarios. Jason Jeyanandan, who set the Foundation up, builds the first ten scenarios people send, in the open. The offer is on the Foundation’s home page: “Give us a real scenario. We will build you a sample.” Received so far: none. Thirty to One is plain about the starting point: “None of it has been measured with real learners. Every experiment so far has had models in the chairs.”

What it means for you

If you fund things, tell us what you would need to see measured first, before any of it runs. The record is not built yet, so it can still be shaped around a question you will actually read. That is where you come in. You can message Jason Jeyanandan on LinkedIn or reply wherever you found this. If you would rather help the work than shape it, send a scenario for the first ten, or support the compute.

This is the last of five. The part before it, How a showcase is made, is the method behind every package; the first, The first run, is the one full run so far, with a simulated learner. The plan above is how a person would replace that stand-in.

What is the first number you would want from us?

Sources

← All writing