Rhetorical Patterns & Detection Cues

This table defines the rhetorical patterns used in the Modular Journalism pipeline. Each Rhetoric captures a recurring editorial tactic that can reduce user value–by misleading, omitting, or distorting information. Beneath each rhetoric, the Detection Cues specify what agents should watch for:

• Lexicons: Specific words or phrases that signal bias, exaggeration, or manipulation (e.g., "shocking," "you won't believe")
• Structures: Formal or grammatical patterns that obscure responsibility or clarity (e.g., passive voice, vague sourcing, missing baselines)
• Guards: Contextual checks that prevent false positives–for example, cases where a phrase is neutral or justified by the surrounding content

Each rhetoric is also linked to User Information Needs–questions that help rebuild the missing or distorted elements. These needs act as editorial countermeasures: they guide agents (or editors) in restoring context, provenance, comparisons, or methods.

In the pipeline, Agent 2 uses these cues to annotate paragraphs. Downstream agents then use the needs to reconstruct modular story units that are transparent, verifiable, and useful.

The pattern and cues dataset is a work in progress and will be tested and finetuned during training.