What changed, and what was corrected

September 4, 2026 · v0.4

One spine

An editorial pass that gives the site one hierarchical argument: when the conclusion arrives before the evidence, that is a sales pitch; when it survives the evidence against it, you are being closed. Every page is now one level of that sentence, and the register was brought into line with it.

Changes

• The homepage tells the whole story in miniature: the same two-option grid three times — a delivery question, the August 2020 sentence, and the checks after the count — with the hinge between the sales version and the big one stated once.
• New: Old names (/lineage/old-names). The moves against the fallacies they already answered to — begging the question, poisoning the well, No true Scotsman, the borrowed kettle — and the documented drift of “beg the question” in English from Fowler to Merriam-Webster, followed by the inoculation studies on naming a technique.
• The canonical sentences (the definition, the small version, the hinge) live in one data file and are imported by the homepage, the engine, the case, About and the glossary; the four rival definitions are gone.
• How it works names the move (petitio principii, cash-or-card, the cart before the horse) before it introduces the site’s term, and its quick test is now the same four questions as everywhere else.
• Outside politics opens one generation earlier, with the 1879–1892 trail of “a sucker born every minute”, and runs the sales curriculum as a chronology to the 2025 order.
• Register: the status of the untested comparisons is stated once, on Method; the pages that depended on it now say what the record supports and point there. About’s preview-build section is retired (the counts are on Evidence).
• Glossary gains Self-sealing, Poisoning the well, No true Scotsman, Kettle logic and the Overton window; the assumptive-close entry is aligned with the hinge.
• “Is this claim closed?” has its own share image, and its two closed verdicts open with the verb.

Corrections

• Outside politics: the delivery-day question is now given as the textbook form of the alternate of choice rather than as J. Douglas Edwards’s verbatim words, which the cited recordings do not contain; Hopkins’ recorded version is quoted alongside it. The archive.org citation now names the album the notes belong to.
• Outside politics: the Amazon case is described as a $2.5 billion settlement rather than a court ruling, and the decline-button wording as a presupposed yes rather than as the assumptive close; the cancellation flow is the alternate of choice.
• Method: the Deetz comparison table called D6 “Retroactive deniability”; the catalog’s name is Retroactive frame control.

September 2, 2026 · v0.3

One definition, one document

A full pass over the site: the argument now uses one definition of closure throughout, the older pages were brought into the same voice as the July revision, and the fact-check corrections below were applied.

Changes

• Canonical definition settled: closure is what happens after an adverse check — the claim is revised, or the checker is reclassified. Presupposition (“the small version”) is now explicitly the entry stage, not a competing definition.
• The five evidence sub-pages were folded into a single Evidence page; old URLs redirect to the matching section.
• New: “Is this claim closed?” (a 60-second check card), “How to talk to someone inside a closed claim”, and this changelog.
• Navigation relabeled (How it works · The 2020 case · The devices · Practice · AI mirror · Method & limits); Glossary and Beyond politics added to the footer.
• Trainer: five new drills so every device and multiplier is covered; the Next button is no longer clickable before answering.
• Fonts are self-hosted; the site sets no cookies and loads no third-party scripts. Share previews, canonical URLs and a sitemap were added, and pages ship a strict content-security policy with nothing inlined.
• “Is this claim closed?” distinguishes an open claim from one that is merely untested, names the unanswered question when a result is requested too early, and links the first question to retroactive frame control (D6), which is the closure device that produces the unpinnable claim.
• Trainer answer order is shuffled on every showing, and the end screen lists the items missed so they can be re-read on the reopen page. Scrollable tables and the interaction matrix are reachable from the keyboard.
• The 2020 case adds Attorney General Barr’s December 1, 2020 statement to the record, dates the SEB2020-059 dismissal (June 20, 2023), and carries a dated note on the January 28, 2026 FBI seizure of Fulton County’s 2020 records, which is treated as an open check, not a finding.
• The lineage page cites the Roman hostis declaration, the Law of 22 Prairial, Khrushchev’s 1956 speech and McCarthy’s February 20, 1950 floor speech to primary or scholarly sources, and the disclaimer now precedes the history rather than following it.

Corrections

• Loftus & Palmer (1974): “triple the false memories of broken glass” corrected to “more than double (32% vs 14%)”.
• Loftus & Zanni (1975): the false-“yes” figures now follow the paper — 15% vs 7% in the first experiment, 20% vs 6% in the second; the 47.5% / 15.5% “don’t know” figures (second experiment) stand.
• Halpern (2013) on default effects: 43% and 77% are the two opposing defaults; the no-default condition was about 61%. The sentence was rewritten to say so.
• Khrushchev’s 1956 secret speech: the paraphrase now follows the text — the “enemy of the people” label made it unnecessary to prove ideological errors, i.e. it removed the accuser’s burden.
• Tydings Committee (1950): the unverifiable “80 names / 42 at State” figures were replaced with the documented record (81 numbered cases, nine named; the committee called the charges “a fraud and a hoax”).
• The Bannon interview attributed on the M3 page is dated to its publication (Bloomberg, Feb 9, 2018).
• Sieker, Lachenmaier & Zarrieß is cited as Proc. CogSci 2025, not Cognitive Science 2025. TruthfulQA is credited to Lin, Hilton & Evans (ACL 2022).
• The D6 dossier no longer claims a “function test” from the timing of the 2016 GRU spearphishing; the retroactive-label asymmetry is kept, the causal implication dropped.
• The lineage page no longer says the historical continuity claim is “operationalized”; it is a testable, untested hypothesis (F2).
• The C-SPAN link on the evidence page pointed at the wrong video id; fixed.
• The D5 note on the August 19, 2020 press briefing now quotes what the press secretary said when asked whether the President would accept the results, instead of characterising the exchange.

August 17, 2026 · v0.2.1

Numbering normalized

Devices D1–D8 and force multipliers M1–M4 were renumbered consistently across the catalog, trainer, matrix and field guide.

Changes

• Device and multiplier numbering aligned across every page and the data files.

July 17, 2026 · v0.2

Clarity revision

The 2020 case page was rebuilt as a claim-by-claim argument with a primary-source ledger; causal claims were tightened; the evidence scaffolds were relabeled as plans rather than results.

Changes

• Case page: verdict box, numbered argument cards, evidence ledger, and an explicit “what would weaken this” section.
• Method page separates the factual record, the rhetorical reading, and uncompleted comparative or causal claims.
• Evidence pages relabeled: the prediction registry, corpus explorer, full-speech check, reopened cases and dataset are described as scaffolds, not findings.
• Mobile accessibility fixes; causal language hedged where the comparison test (F2) has not run.

Corrections

• The prediction registry entries were previously labeled “posted July 6”; the repository holds no timestamp supporting that. They are now dated to their first capture in version control (July 17, 2026) and marked as drafts.

July 17, 2026 · v0.1

First public version

The catalog of twelve constructions, the seven-stage map, the trainer, the map-it tool, the AI mirror, and the lineage pages.

Changes

• Initial release.