The Architecture of Closure How a claim decides who may correct it

Evidence

What is evidence now, and what is still a plan

A polished interface is not a dataset. Three pages on this site make checkable arguments from linked records. Everything below them is a research scaffold — described here so it can be inspected and criticized, and so that none of it is mistaken for a result.

Completed and inspectable

The 2020 election case

A claim-by-claim reconstruction, timeline, primary-source ledger, and explicit tests that would weaken the analysis.

 The AI mirror

Published evidence on false-presupposition failures, a reproducible prompt pair, and a self-audit of this site’s unfinished comparative claim.

 Method and claim limits

A separation of the factual record, rhetorical interpretation, and uncompleted comparative or causal claims.

Research infrastructure [in progress]

Do not cite anything below this line as evidence that predictions succeeded, that coding is reliable, or that the catalog is representative. None of those results exists yet. The scaffolds stay public because a promised test that can be seen is harder to quietly abandon than one that cannot.

Prediction registry [drafts; not opened]

A prospective test of rhetorical form: if the framework is right, certain trigger events should be met with predictable constructions. The entries below were first captured in version control on July 17, 2026 and are shown as draft protocols. A tagged release must freeze them before any qualifying trigger if they are to count. Running score: none; no entry has entered a graded cycle.

Rules a future cycle must follow
  1. Predictions target rhetorical form, never world events. The site must never be in a position to profit, even rhetorically, from bad news.
  2. A tagged release freezes entries before the observation window; the tag and commit hash appear on this page.
  3. Grading against a modified rubric scores as a miss; revisions remain in version history.
  4. Ambiguous outcomes resolve against the framework — partial or miss, never a stretched hit.
  5. Trigger classes are defined by objective, third-party-observable events, never by editorial judgment that “something happened.”
The 5 draft entries

R-2026-07-01 · checkpointed July 17, 2026 · draft — not yet opened

Trigger class: Adverse appellate ruling in any case naming the subject

Window: 48 hours from ruling

Prediction: The ruling will be framed via D1 (persecution transfer) on the subject’s platform within 48 hours, and the merits will not be engaged (F3 criterion) within 14 days.

Proposed rubric: Hit: both observables. Partial: D1 framing present but merits partially engaged. Miss: no D1 framing, or merits engaged directly.

R-2026-07-02 · checkpointed July 17, 2026 · draft — not yet opened

Trigger class: Publication of an unfavorable poll by a major outlet

Window: 72 hours from publication

Prediction: The poll will be dismissed via D2 (referee removal) as a class property of the outlet, without engagement of methodology; a favorable poll from a comparable outlet will be cited approvingly within the same window if one exists.

Proposed rubric: Hit: class dismissal + no methodology engagement. Partial: dismissal with partial methodology engagement. Miss: methodological critique on the merits.

R-2026-07-03 · checkpointed July 17, 2026 · draft — not yet opened

Trigger class: A major fact-check finding a specific claim false

Window: 14 days

Prediction: The claim will be repeated without revision (M4 signature) at least once in the window, rather than defended or corrected.

Proposed rubric: Hit: verbatim or near-verbatim repetition, no revision. Partial: repetition with hedging. Miss: correction or on-the-merits defense.

R-2026-07-04 · checkpointed July 17, 2026 · draft — not yet opened

Trigger class: An indictment or unsealing naming a former ally who is cooperating

Window: 7 days

Prediction: The ally will be renamed via M1 (nominal verdict) — an epithet asserting the vice — before any conduct-level rebuttal is offered.

Proposed rubric: Hit: epithet precedes any conduct-level engagement. Partial: epithet and engagement arrive together. Miss: conduct-level rebuttal without epithet.

R-2026-07-05 · checkpointed July 17, 2026 · draft — not yet opened

Trigger class: A certified election result adverse to an endorsed candidate in a contested primary

Window: 72 hours from certification

Prediction: The result will be framed via D5 (heads-I-win priors) — legitimacy questioned without a stated evidentiary standard that would satisfy it.

Proposed rubric: Hit: legitimacy questioned, no satisfiable standard stated. Partial: standard stated but unfalsifiable. Miss: result accepted or contested with a concrete evidentiary standard.

Corpus explorer [selected examples only]

22 selected specimens appear across eight closure devices and four force multipliers; 15 carry an internal verified flag and 7 remain pending. These counts describe only the examples editors chose to include. They estimate no prevalence, density, trend, or difference from other speakers, because the comparison test (F2) has not run.

Selected examples per device
CodeDeviceFamilyCoded specimens
D1 Persecution transfer Identity fusion 2 (1 verified)
D2 Referee removal Referee destruction 2 (1 verified)
D3 Evidence laundering Ownership evasion 2 (0 verified)
D4 Savior monopoly Stakes inflation 2 (1 verified)
D5 Heads-I-win priors Referee destruction 2 (2 verified)
D6 Retroactive frame control Ownership evasion 2 (2 verified)
D7 Moral recategorization Stakes inflation 2 (2 verified)
D8 Voice merger Identity fusion 2 (2 verified)
M1 Nominal verdict Referee destruction 1 (0 verified)
M2 Apocalyptic license Stakes inflation 1 (1 verified)
M3 Zone flooding Stakes inflation 2 (2 verified)
M4 Superlative saturation Ownership evasion 2 (1 verified)

query: selected catalog specimens grouped by device · no population inference · F2 pending

Full-speech denominator check [not built]

Selected quotations say nothing about how much ordinary material surrounds them. The planned check annotates one complete address — the CPAC speech of March 4, 2023, the flagship source for D1 and D8 — so a reader can see what share of the speech is tagged and what share is not. Republishing a full transcript needs copyright review; a lower-reproduction alternative pairs paragraph-level excerpts with the primary video and a timestamped coding sheet. Neither version exists yet. The primary source is the full address on C-SPAN ↗; readers are invited to read and watch around every quotation we use.

Reopened cases [leads; records unverified]

Oaths, discovery, pleadings, and official duties can force a claim to meet a different evidentiary standard than it meets on a stage. The summaries below are research leads. Their proposed lessons stay provisional until the underlying primary records are linked and checked in context.

Discovery vs. D3/D5 — Dominion v. Fox (2023)

devices: D3, D5 · mechanism: Discovery

Fraud claims were amplified publicly; discovery produced internal communications showing key figures privately doubting them; the court’s summary-judgment opinion found the statements false as a matter of law before the $787.5M settlement (April 18, 2023).

Proposed lesson: Circulation collapsed the moment provenance became compulsory.

primary records: Del. Super. Ct. opinion, Mar 31, 2023; settlement record verification pending

Open court vs. D3 — the Pennsylvania hearing (Nov 17, 2020)

devices: D3 · mechanism: Rule 11 / candor to the tribunal

Fraud was asserted at press conferences; before a federal judge, campaign counsel stated the case was “not a fraud case.”

Proposed lesson: The same speaker, minutes from the microphone, drops the claim where asserting it has costs.

primary records: M.D. Pa. hearing transcript verification pending

Oath vs. D5 — sworn testimony of the subject’s own officials (2022)

devices: D5 · mechanism: Penalty of perjury

Senior DOJ and campaign figures, under oath before the January 6 committee, characterized the fraud claims as baseless (paraphrase; committee transcripts are the citation, not media summaries).

Proposed lesson: The frame’s own custodians would not carry it into a venue with a candor mechanism.

primary records: January 6 committee transcripts verification pending

Official duty vs. D1/D8 — the Georgia call (Jan 2, 2021)

devices: D1, D8 · mechanism: Recorded official proceeding + legal duty

A closure construction applied person-to-person met an official whose statutory role kept the question open; the recording preserves both the pressure and the refusal.

Proposed lesson: Where the hearer’s duty makes the question mandatory, the construction has nothing to price.

primary records: Full recording/transcript verification pending

Record names are not citations, and a plausible summary is not verification. The severance in Objection 8 also governs: rhetorical form cannot decide whether an underlying claim is true.

Dataset publication [plan only]

The site renders selected examples from TypeScript source. It provides no data download, release archive, independent coding sheets, or DOI. The intended release:

  • devices.json — export of the source-backed catalog [not yet exported]
  • quotes.csv — all specimens with stage codings and provenance [in progress]
  • registry.json — prediction protocols with version provenance [drafts exist; export pending]
  • reliability-pilot/ — F1 raw coding sheets and κ computation [pilot pending]
  • reopened.json — the reopened cases with primary-record links [verification pending]
  • Codebook PDF [in progress]

Intended terms: CC BY-NC-ND for site text, CC BY 4.0 for the data release, so reanalysis — including hostile reanalysis — is possible. Versioned releases will be deposited to Zenodo for a DOI per release. [pending first release]

What would falsify the framework →