Frozen Set 1

A freeze is not a pause in the work. It is a fixed target: for one round nothing published here changes, every reader audits the same text, all findings arrive, and a single revision wave applies them. Without it each correction invalidates the audit that produced it and letters cross in the post — which is what happened through the night of 1 September, when three reviewers were each reading a different snapshot of the same documents.

Frozen 2 September 2026, 02:38 local (23:38 UTC), at commit ad6e97f. The working tree was clean, and every document below was verified live by cache-busted fetch at the moment of freezing — because a fetch that could be served from cache is not a reading of the live site, and two of the three audits that night were misled by exactly that.

The set

Last scan: 2026-09-01 23:39 UTC · cache-busted fetch · 9 of 9 in sync

The book, The Puppet Condition: Restrung, is frozen at the third assembled draft and is not public. It draws on every document above, which is why a change to any of them obliges a pass over it.

The protocol’s deposit at osf.io/ceauh carries the text as it stood at 22:11 UTC on 1 September. Section 1 of that page states precisely what the deposit covers and what has changed since: sections 2 through 9 — every hypothesis, threshold, decision rule, exclusion and procedure — are identical, and the differences are confined to the header, that section, and the changelog’s presentation. The next deposit carries the current state.

What the freeze covers, narrowed 2 September

The freeze is on the text, not on its presentation. The audits under way are reading claims, thresholds and rules; none of them is reading a stylesheet. Holding the site’s appearance hostage to a text review would stop work that no reviewer is waiting on, for no gain to the review.

So: wording, structure, numbers and links are frozen — every sentence a reader could audit stays exactly as it is until the wave. Typography, layout, colour, imagery and navigation are not, and may change while the set is frozen.

One consequence, stated because it is the obvious way to cheat this: a redesign that adds, removes or rewrites any sentence is a text change and waits for the wave, whatever it is called. The scanner enforces the line mechanically — its canaries are strings from the frozen text, so any presentation change that disturbs a sentence fails the scan on the next run.

The set was opened once, on 2 September

Three documents in this set stated that four preprints were under review at academic journals. All four had been returned rejected. A freeze protects a review from moving text underneath it; it does not license continuing to publish something the Institute knows to be false. A factual error of this class is corrected on sight, and the rule that holds errors for the wave covers matters of wording, structure and emphasis — not claims about the world that have stopped being true.

So the set was opened, the claim corrected everywhere it appeared, and closed again. It is recorded here because a freeze broken quietly is not a freeze:

No reason for the rejections is stated in any of them, because the Institute does not have one to state and a supplied reason would be commentary rather than record. Nothing else moved, and every item collected for the revision wave is still waiting for the wave.

And a second time, on 2 September, for a decision rather than an error

The first opening corrected something false. This one did not: the Founder decided that the four papers extracted from the first edition should not be carried on this site at all, and the pages were deleted.

That is a decision, not a correction, and the difference is written here rather than blurred. A freeze can be opened for a factual error on sight; it should not be opened for a preference. It was opened anyway, because the alternative was leaving two documents pointing at pages that no longer exist, and a dead link is a smaller lie than a live one but it is still one.

What moved, and only this:

Both changes are deletions. Nothing was reworded and nothing was added, and that is deliberate: a deletion cannot introduce a claim, so an opening made only of deletions cannot smuggle a change in behind a correction. The same rule now governs errata in the canon, for the same reason.

The papers themselves are unaffected. Each carries its own DOI on Zenodo and is indexed on PhilPapers; the site simply no longer hosts them. The account of why the second version waited — that four papers were under review, and what became of them — stands in the Records exactly as it was.

A third time, and what three openings in one day mean

The waitlist on /theoi/ was removed, together with the paragraph that pointed at it. The form now has its own page and its own place in the bar, so the same address was being collected in two places with the second one buried inside a section; the Founder asked for one.

Again both edits are deletions: the form went, and the sentence that said “leave an address below” went with it, because deleting only the form would have left a paragraph promising a field that was not there. Nothing was reworded and nothing was added. What remains of the section stands on its own and says nothing that is not true.

Three openings in one day is a fact about the freeze rather than about the changes. The first was a factual error and belonged to the exemption. The second and third were decisions, and a freeze should not be opened for a decision.

Frozen Set 1 was declared for one audit round. That round has closed — the review letters are logged, the last reviewer has recused, and the closing document is written. The freeze page says what comes next: one revision wave, then release. Until that wave runs, the set is no longer protecting a review from moving text; it is holding twelve collected items behind a discipline whose purpose has been served, and the deletions above are what it looks like when that starts to cost more than it saves.

Released, 2 September 2026

The wave has landed and Frozen Set 1 is released. The round it was declared for had closed — the letters are logged, the last reviewer has recused, the closing document is written — and past that point the freeze was not protecting a review from moving text. It was holding twelve collected items behind a discipline whose purpose had been served, which is the state the three openings above were already symptoms of.

What the wave changed, by document:

Section 1 is the entry that matters most, because the wave broke it. That section said the deposited text and the live text were identical in every clause governing the trials, and the wave changed Sections 4 and 6. The replacement does not restate the reassuring claim in weaker words: it names which sections still match the deposit, which two do not, and what kind of change each was — in both cases an addition fulfilling what the deposit promised rather than a threshold moved after the promise. Until the next deposit is made, the page says plainly that it is ahead of its own timestamp.

That is the one thing a revision wave can most easily get wrong. A page that describes its own relationship to a deposit will be made false by the very edits the wave applies, and it will be made false quietly, in a paragraph nobody re-reads because it was not on the list. It was on this list because it was item 6.

One item was withdrawn rather than deferred. The wave was to end with a second OSF deposit carrying the current state. Put to the Founder, it did not survive the question what is this deposit for — and the answer is that it is for nothing yet. No data exists; the first trial follows the close of the first Age. The 1 September deposit already timestamps every clause that governs the trials, and this document is in the regime its own Section 10 calls free and frequent revision, so a deposit now would permanently record a state its successor supersedes. One deposit, when the protocol is final, before the first trial. Section 1 says so with its date rather than leaving an open promise.

What happens now

The set is released and the work resumes. No new set is declared: the next round, if one is needed, gets its own freeze, its own commit hash, and its own cache-busted verification — and, on this round’s evidence, a stated end date, because a freeze that outlives its round stops protecting anything and starts costing.