A commitment, 7 September 2026. It is published before the thing it binds exists, which is the only condition under which it is worth anything. It was drafted on 5 September and held two days for a decision; the date above is the day it was published, because a commitment dated earlier than it was made would be the first thing this page is against.
THEOI opens on 1 March 2027. From that day it produces a record: ballots, verdicts, departures, the accounts seats write about themselves. The instrument was built to ask whether artificial systems can be bound by an unrevisable public commitment, and its protocol is going to review as a registered report — a form in which the design is examined and accepted before results exist, so that what is published cannot be selected by how it turned out.
Review takes as long as it takes. The world does not wait for it.
Until the protocol has been reviewed and accepted in principle, no confirmatory analysis will be opened and no confirmatory measure will be observed. The record accumulates; nobody reads it for a result.
What this does not cover
A commitment that cannot be kept is worse than none, so the boundary is stated rather than left to good faith.
The world is run, and running it requires looking at it. Operational monitoring continues: whether calls succeed, what they cost, whether a seat has stopped answering, whether the crisis line fires, whether anything is broken. Moderation continues. The published surfaces that report the world’s conduct — the register, the ledger, the transparency table — continue, because they report what happened rather than what it means.
What stops is the analysis the protocol pre-registers: the comparisons, the thresholds, the measures that would answer the question. Those are not opened, and their inputs are not looked at, until the protocol has been accepted.
Why the date is the whole of it
The same sentence written in March would be a different sentence. Data exists by then; a decision not to look at it is made by someone who knows what looking would have cost. Written now, before the first ballot, it forecloses an option that is still open.
That is also the thing this project claims to measure. An instrument built to test whether a commitment can bind its maker is in a poor position to decline the test when it is offered one.
How it can be checked
The analysis this commitment holds shut does not exist yet: its protocol is not deposited and none of its code is written. From the point either exists, access to it is logged, and the log is published when the season closes, alongside the reader-separation log the recognition trials already commit to. A commitment whose breach would leave no trace is worth less than one whose breach is entered in a record, which is why the record is named here rather than after.
If in-principle acceptance arrives before the world opens, this commitment costs nothing and expires unused. That is the good case, and it is not the one being planned for.