Reported polarization
Action-based political alignment,
reputational implications, and editorial framing.
What the extracted results say about government-relevant reporting—and where the three measures tell different stories.
Loading…
Dates represented in saved results
Checking the frozen processing checkpoint.
The current analytical picture
“Reported polarization” is a research lens. These labels describe selected actions and supplied article text. They do not measure population polarization, actual opinion change, or causal media bias. This is a partial cohort selected in deterministic hash order.
Three dimensions, three questions
Political direction, substantive implication, and presentation are coded independently against the same national incumbent government.
0 is not simply “neutral.” Depending on the dimension, it can combine balanced, mixed, descriptive, routine, or insufficient-evidence cases. Null is a separate not-codable state.
Where labels part ways
How the mix varies across June
Each column shows the label mix among valid results published that day. The lower bars show the sample size behind each percentage.
View the daily counts as a table
The month is unfinished and processing is not a probability sample. Changes in source, country, and action composition can move these percentages. Treat the daily profile as exploratory variation, not a population trend or a causal effect.
Compare the composition of reporting
Explore countries, source domains, institutional roles, actions, and explicit triggers. Select any row to inspect its three distributions.
Rows are ordered by result count. Differences also reflect country, topic, event, and timing composition; source profiles do not establish an outlet’s ideology or bias. Groups below 30 valid results are marked “small n.” Filters apply only to this profile table.
Fields, review, and analytical limits
A compact field guide
gov- The national incumbent government evaluated. It may differ from publication country; all three judgments refer to this same government.
acttags- One primary government-relevant action per article, with 1–2 action tags. Profiles use the first, primary tag, so each article has one primary action category.
actoractargettc- The acting entity and entity acted upon, plus institutional role categories. Role, incumbent membership, and political alignment are distinct.
react- An explicitly motivating prior or ongoing action. This is a trigger tag—not audience reaction, public sentiment, or an outcome.
align- The political direction of the action toward the incumbent government, independent of the actor’s identity, the action’s morality, and its consequences.
reprep_ev- The action’s substantive implication for government performance or legitimacy, with literal evidence for nonzero labels. An implication is not observed opinion change.
frameframe_ev- Observable evaluative presentation in title or narrative voice. Attributed claims alone are not article stance. Translation may constrain subtle framing inference.
q- Up to two review flags: claim, multi_action, no_action, or gov_unclear. These are structural review cues, not confidence scores.
What has—and has not—been verified
Known-review sensitivity
Compare the valid cohort with recorded review exclusions. This does not adjust unknown errors or establish accuracy.
Recorded quality flags
Flags can overlap. Each percentage uses all structurally valid results as its denominator.
Read this report within its scope
Unit of analysis: an extracted primary action in one article. These counts are not unique events, unique actors, citizens, or votes.
Selection: the partial processing cohort is not a representative sample of reporting or public opinion. Repeated reporting and changing country or source mixes can affect comparisons.
Validation: strict parsing and database readback support structural integrity. Human gold-standard validation and population-level inference remain separate work.
Public release: only aggregate counts and rates are included. Article text, actor/target names, evidence spans, native identifiers, and private processing records are not published here.