Skip to main content

​
Signals

A signal is a market intelligence article that has been ingested, classified, and decomposed into independently verified text units.

​
Source types

Signals are classified into two content types during ingestion:
TypeDecompositionExample
analysisThesis + arguments + claimsResearch reports, opinion pieces
eventSummary + claimsNews articles, press releases

​
Decomposition structure

​
Analysis signals

Signal
├── Thesis (1:1)
│   └── Thesis sentences (verified)
└── Arguments (1:N)
    ├── Argument sentences (verified)
    └── Claims (verified)
The thesis is the article’s core argument. It’s decomposed into individual sentences, each verified against source evidence. Arguments support the thesis — each has its own sentences and claims.

​
Event signals

Signal
├── Summary (1:1)
│   └── Summary sentences (verified)
└── Claims (verified)
The summary describes the event. Claims are specific factual assertions extracted from the article.

​
Text units

All decomposed pieces — thesis sentences, argument sentences, summary sentences, and claims — are text units. This is the fundamental atom of Gildea’s data model. Every text unit has:
  • A verification verdict (only pass is served)
  • A scoring mode (nli for claims, regression for sentences)
  • A primary score (method-specific — do not normalize across modes)
Claims and sentences use different scoring methods. Do not compare primary scores between nli and regression modes.

​
Signal card vs. signal detail

The list endpoint returns lightweight signal cards with metadata and a decomposition_text preview (thesis or summary text). The detail endpoint returns the full decomposition tree with all verified text units. Use include=evidence for source evidence snippets. See Progressive Disclosure for the full pattern.