POWERED BY TYPESAFE / JEV
A model that
makes a choice.
Jev is a structured decision model. Give it context, a question and defined options; it returns a choice and the probability of each option.
Jevlings uses typesafe/jev-1.13 to explore a narrative through six audience perspectives. The characters make those perspectives visible. Jev supplies the judgments behind them.
FROM STORY TO CROWD
The Jevlings pipeline
- 01
Collect & review
Start with a written story, or resolve a Solana token address into metadata and public project materials. The research service extracts a narrative; review and edit it before analysis.
- 02
Frame six perspectives
Pair the same narrative with lenses for meme culture, novelty, community, creativity, skepticism and accessibility. Each lens has the same eight reaction choices.
- 03
Ask Jev
Batched requests obtain reaction distributions, then supporting reasons and relevance judgments. The service validates the returned options and probabilities before saving a completed report.
- 04
Simulate 1,000 reactions
A seeded sampler turns the six distributions into 1,000 reactions. Every visible map node represents one simulated reaction, and the report uses those same totals.
THE MODEL INTERFACE
Context in. Typed decisions out.
The server sends the narrative in state and perspective-specific prompts in questions. Each answer contains a selected choice, a probabilities object and confidence.
{
model: "typesafe/jev-1.13",
state: { narrative: "…" },
questions: {
moss: { /* perspective + choices */ },
/* five more perspectives */
}
}- choice
- The selected reaction.
- probabilities
- A value from 0 to 1 for each reaction option.
- confidence
- The model’s reported confidence.
Eight options: passed, read, liked, questioned, shared, followed, muted and unsure. The interface labels questioned as “Disagreed.” Invalid or incomplete distributions do not become a result.
THE APPLICATION LAYER
From probabilities to a result
Jev returns probabilities. Jevlings samples the crowd and calculates its own Narrative Response Index.
Liked, shared and followed count as positive. Disagreed and muted count as negative. The other reactions are neutral. The scale runs from 20 to 100; an even balance scores 60.
The map is a modeled projection, not 1,000 independent model calls or human respondents. Use it to compare interpretations of a story; it is not a token price forecast or a measured audience outcome.
Example and earlier research
300 positive and 200 negative reactions give an index of 64. Replaying the map keeps the saved totals. Earlier archived research used a different rubric and cannot be compared on this scale.
Open the research archive →