The Rhetoric League

Checklist preview · chk-v1 · recomputed every build · as of 2026-07-21

How to read this — four things, before the numbers

This is a preview of the deterministic checklist layer only — five rule detectors counting a fixed set of language features. It is not a verdict on any newsroom.
It is reported at outlet-type level because no individual outlet yet clears the minimum-sample gate (n ≥ 30). A type pools every scored article filed under it.
The Aristotle appeal triangle, inter-coder reliability (Cohen’s κ on a human-coded gold set), and bootstrap confidence intervals are all pending. With no intervals yet, no cell here should be read as ranked above another — the differences are not tested.
Every count is span-grounded to a character offset in a frozen snapshot and was R4-verified at 0% offset drop. The receipts below resolve each detector hit back to the exact verbatim text.

3×3outlet types by checklist items that clear the sample gate. Each figure is a rate — instances the rule detectors counted per 1,000 words — never a charge against the outlets pooled into it.

Most of the desk’s beat is the gap between two accounts of one event. This rail is narrower and more mechanical: it counts language features — passive constructions with the actor deleted, hedges, euphemisms — with deterministic rules, then pools the counts by outlet type. The rules fire the same way every run; the number is a count per 1,000 words, nothing more. What a reader does with a rate is their own; the desk only logs it.

The league

Pooled rate per 1,000 words (2 decimals) with n = scored articles, by outlet type. A cell shows only at n ≥ 30.
Outlet type	AGPagentless passive	EUPeuphemism	HDGhedging
national	0.90n=90	0.13n=90	3.48n=90
cable	0.38n=37	0.08n=37	4.36n=37
wire	0.12n=39	0.00n=39	1.61n=39

Held — below min-n (n<30): items ATL (attribution-laundering), VBI (harm-verb intensity); outlet types primary source, digital, opinion heavy. These are not shown until they clear the sample gate.

Rows and columns are in a fixed order, not a ranking. Without confidence intervals, no ordering among these cells is significant — a higher rate is a higher count, not a finding that one type differs from another.

What the columns count

AGP agentless passive — passive voice with the actor deleted — the empty chair
HDG hedging
EUP euphemism
ATL attribution-laundering
VBI harm-verb intensity

The receipts

Two instances per shown item, pulled from the frozen bodies at their exact offsets — the verbatim span marked, the outlet and character range beside it. This is what “every count resolves to a char offset” means in practice.

AGP · agentless passive

…rd to arrest someone other than the man who was shot. The brief DHS statement did not menti…

Al Jazeera (national) chars 1010–1022

…son of U.S. District Judge Esther Salas who was shot and killed at their New Jersey home by …

CBS News (national) chars 2889–2901

EUP · euphemism

…ruck more than 80 Hezbollah targets and eliminated dozens of Hezbollah terrorists after fo…

Fox News (cable) chars 2872–2882

…y, July 11, U.S. Senator Lindsey Graham passed away from a brief and sudden illness," his o…

The Guardian (national) chars 446–457

HDG · hedging

… offered little evidence to support its claims of killing "narcoterrorists." The stri…

CBS News (national) chars 1395–1401

…es during its war with Iran, after some denied the U.S. basing and overflight rights f…

Reuters (wire) chars 952–958

[OUTPUT] Deterministic checklist layer (chk-v1): rule-detector language features counted per 1,000 words, each span-grounded to a char offset in a frozen snapshot (R4 offset-resolution: 0% drop). Pooled at outlet-type level over 212 scored articles; cells shown at n ≥ 30. Appeal triangle, κ, and confidence intervals pending — rates are counts, not a ranking. probability mass ≠ 1.0.