JEV-INSPIRED TYPED DECISIONS

One state. Parallel questions.
Typed answers.

MEDJEV adapts the JEV idea to biomedical evidence: send one shared state, ask bounded questions at runtime, and return structured decisions with explicit probabilities instead of generated prose.

MEDJEV logo
JEV decision loop
Ready · same state, three question types

One state, many bounded judgments

JEV evaluates several typed questions independently against the same state; adding a question does not turn the answer into generated text.

01 · STATE

Shared input context

Clinical record +
biomedical passage
encoded once
02 · QUESTIONS

Typed questions

CHOICEWhich evidence relation?
supported · contradicted · unresolved
SCOREHow strong is the signal?
ordered rubric · 1 · 2 · 3 · 4 · 5
NOULIs the statement supported?
yes / no · probability of yes
03 · ANSWERS

Decisions software can use

supported
0.74
3.8 / 5
0.81
yes
0.68

The output is a typed decision and a probability distribution—not a generated explanation.

ChoiceSelect one item from the candidate set defined by the caller.
ScorePlace the state on an ordered rubric and preserve level probabilities.
NoulReturn the probability of a bounded yes/no judgment.