Method
The codebook, and what would prove us wrong
If this framework can’t fail, it’s an instance of the thing it describes. Four tests below could kill it. One runs before v1.0 ships. The other three are standing invitations, and the intake form is real.
Corpus
What counts as data:
- Primary: campaign rally transcripts and video (C-SPAN; Roll Call/Factba.se archive); official remarks (The American Presidency Project, UCSB); the archived Twitter corpus (Trump Twitter Archive V2); Truth Social posts (archived captures); debate transcripts (CPD/UCSB); The Art of the Deal (1987) for the “truthful hyperbole” doctrine.
- Secondary: the scholarly literature on the Sources page.
- Fact-adjudication: court filings and rulings, bipartisan official findings, and at least two independent fact-checking organizations for any contested factual characterization.
The codebook
Operational criteria for identifying each stage in a transcript. Inclusion and exclusion rules, one paragraph per stage:
| Stage | Include when… | Exclude when… |
|---|---|---|
| Surface claim | A defensible literal reading exists that the speaker could restate under challenge. | No coherent literal reading exists (pure exclamation, crowd work). |
| Presupposed frame | Parsing the sentence requires accepting a contestable proposition not asserted by it. | The “presupposition” is explicitly asserted elsewhere in the same passage (then it is a claim, not an installation). |
| Hidden warrant | The conclusion follows from the grounds only via an unstated general rule a hearer must supply. | The rule is stated and defended in the passage. |
| Forbidden question | A specific empirical or moral question, if asked in the utterance’s context, would mark the asker as enemy, dupe, or bore — and the corpus shows the question unengaged. | The question is engaged on the merits anywhere proximate in the corpus (this is the F3 check). |
| Emotional cue | Affect-language or delivery whose function is to raise the social cost of the forbidden question. | Affect incidental to the construction (ordinary applause lines). |
| Permitted/excluded conclusions | The passage leaves exactly one low-cost exit, and a specific ordinary conclusion has been made costly. | Multiple exits remain live at comparable cost. |
Boundary cases for each stage — two per stage, as the full codebook requires — ship with the dataset. [in progress]
Falsifiability
Here is what would count against us:
| Test | Failure condition | Status | |
|---|---|---|---|
| F1 | Reliability. Independent coders use the codebook on a stratified sample of 60 utterances (top-4 specimens per device drawn blind, plus 12 distractors containing no closure construction — the distractors keep the instrument honest about false positives). | Cohen’s κ < 0.7 on device identification → constructs are ill-defined; the catalog is revised or withdrawn. Revision history stays public. | pilot pending — required before v1.0 |
| F2 | Baseline. Device density and co-occurrence compared against matched rally/stump corpora of at least three other presidential-level figures (suggested: Obama 2012, G.W. Bush 2004, Biden 2020; same venue type, same word counts). | Densities statistically indistinguishable from baselines → the systemic claim fails (individual dossiers may survive as instances). | not yet run — this is an open invitation |
| F3 | The forbidden-question test. Does the corpus show the speaker regularly engaging the mapped forbidden questions directly and on the merits? | If yes, the corresponding dossiers are wrong and get struck. | not yet run — this is an open invitation |
| F4 | Prospective classification. New utterances should be classifiable by third parties using the codebook without post-hoc category stretching. | Documented failures accumulate on this page. | open — failures ledger empty so far |
Adversarial collaboration
A standing, structured invitation, on a monthly window aligned with the registry cadence. Bring a claimed counterexample or a hostile coding exercise, with primary sources. We pre-commit to publishing outcomes within 30 days regardless of result — including “they were right; we struck or revised the entry.” Capacity is capped at two submissions per cycle. Cranks are filtered by the form’s evidentiary requirements, not by editorial gatekeeping — the structure does the moderating.
First-cycle target: one skeptical academic or practitioner codes a transcript sample with the codebook; the comparison is published either way.
The Deetz mapping
The site’s title concept is Stanley Deetz’s (1992): discursive closure exists “whenever potential conflict is suppressed” (Democracy in an Age of Corporate Colonization, ch. 7, deriving from Habermas’s systematically distorted communication). The engine maps onto his canonical strategies:
| Engine stage | Primary Deetz strategies engaged |
|---|---|
| Presupposed frame | Naturalization, neutralization |
| Hidden warrant | Legitimation, naturalization |
| Forbidden question | Topical avoidance, disqualification |
| Retroactive deniability (D7) | Meaning denial / plausible deniability |
| “That’s just their opinion / everyone’s corrupt” | Subjectification, pacification |
Exact page citations to Deetz ch. 7 are being confirmed against the print edition. [verification pending]
Confidence-marking conventions
Sitewide, claims whose status matters carry small bracketed labels in monospace: [settled], [contested], [illustrative model], [our coinage]. The backfire-effect section on the Mind page and the loop model on the System page are the exemplars. “Forbidden question” and “heads-I-win prior” are this site’s coinages [our coinage]; they should not be attributed to the literature.