Social Monitor
A multi-source news monitor assesses reader value using a structured Jev rubric.
Added to Jevfast
How it uses Jev
The service builds a bounded request from a trusted user interest, a captured title and source text, and capture-availability context. Jev is asked to classify usefulness, relevance, context sufficiency, and evidence basis. The shown code constructs these questions and sends the input to the scorer; the available evidence does not establish a public-facing filtering rule or downstream action for an individual assessment.
What Jev decides
Reader-value assessment request shapeExample answers · not a recorded Jev response · Source ↗
Question 1 · usefulness
YOUR APP
INSTRUCTION
Classify concrete usefulness of the title and captured text for the trusted interest, without assuming unseen content.
STATE
Trusted interest, title (bounded to 2,000 characters), bounded captured source text, and context-state availability are combined into one input. Source text is treated as untrusted. The request uses the four rubric questions below.
JEV · CHOICE
- noise
- context
- useful
- important
- insufficient_context
Question 2 · relevance
YOUR APP
INSTRUCTION
Classify how directly the title and captured text relate to the trusted interest.
STATE
Trusted interest, title (bounded to 2,000 characters), bounded captured source text, and context-state availability are combined into one input. Source text is treated as untrusted. The request uses the four rubric questions below.
JEV · CHOICE
- unrelated
- adjacent
- relevant
- central
- insufficient_context
Question 3 · context sufficiency
YOUR APP
INSTRUCTION
Judge whether the available title and text suffice to assess the contribution without opening a link.
STATE
Trusted interest, title (bounded to 2,000 characters), bounded captured source text, and context-state availability are combined into one input. Source text is treated as untrusted. The request uses the four rubric questions below.
JEV · CHOICE
- insufficient
- partial
- sufficient
Question 4 · evidence basis
YOUR APP
INSTRUCTION
Characterize the support visible in the title and captured text.
STATE
Trusted interest, title (bounded to 2,000 characters), bounded captured source text, and context-state availability are combined into one input. Source text is treated as untrusted. The request uses the four rubric questions below.
JEV · CHOICE
- observation
- described_data
- linked_claim
- unsupported_claim
- no_claim
- insufficient_context
App workflow
- Capture source context
Capture title, source text, trusted interest, and availability context under the source-content safety policy.
- Build assessment request
Bound the title and source text to fit configured request-size limits, then include the four versioned rubric questions.
- Score and persist
The assessment use case dispatches the request to its scorer and persists the result. The inspected evidence does not specify a user-visible filtering threshold or per-item action.
Why it is interesting
The rubric separates whether an item is on-topic from whether its available text supports a useful, sufficiently evidenced contribution. It explicitly guards against treating promotion, engagement, or an unseen link as proof of value.