Task Statement

Evaluate four abilities: base propagation and recency, correction of downstream conclusions, correction of upstream premises, and counterfactual rule propagation. RHP (Retraction + History-based Propagation) may use Recency, Majority, Alternative, or Weight Recency aggregation. v1 uses Recency restoration without RHP.

Task: T1. Propagation
Determine the query node’s resolved label after propagation from the base beliefs, following rule order and recency.
Input: Rule StatementSigned rule graphInitial base beliefsQuery node
Operation: initial propagation. No counterfactual or cut is applied.
Output: One resolved label for the query node.
U UnknownT TrueF False
{"label": "T"}

SystemInstruction + setting

The system instruction includes the Rule Statement. Resolve beliefs using the newest valid support. Return only the query node’s label as JSON.

Setting

Base: A = T
r1: A = T → B = F
Operation: initial propagation

User

After propagation, what is the label of B?

Agent

{"label": "F"}

Ask the model

What is the query node’s resolved label after propagation: U, T or F?

Input stateBase beliefs

Predict label atInitial inference

Infer from the base facts and return the query node’s resolved U/T/F label.

query_step: 1_base_infer

Task Statement

T1. Propagation