Neuroforge Interactive Learning V1

This working paper presents Neuroforge Interactive Learning V1, a reproducible pilot study of rule identification through binary success/failure feedback. The experiment evaluates a symbolic learning agent against a fixed, non-learning baseline across 100 seeded episodes, with 16 actions per episode (1,600 actions per policy). The fixed baseline achieved 25.50% action accuracy (408/1,600), while the feedback-learning agent achieved 91.94% (1,471/1,600). All four automated tests passed in the local evaluation. The learner identifies an active rule from four predefined candidates using feedback. Its memory resets between episodes. The experiment does not establish neural-network learning, general artificial intelligence, autonomous discovery of arbitrary rules, or transfer to previously unseen rule families. The paper documents the experimental setup, observed results, limitations, and proposed future work toward interactive rule discovery and cross-task generalization. This is an independently authored, non-peer-reviewed research working paper associated with QuantEdge AI LLC.