## Overview.
**ProcessVenue’s** internal annotation team created a sarcasm detection dataset consisting of 5,940 real-world sentences, labelled under a strict double-blind dual-annotation strategy. The task focused on accurately separating Sarcastic text (ironic contradiction between literal meaning and intent) from non-sarcastic text (direct meaning).
Despite the inherently subjective and context-light nature of sarcasm, the project achieved strong individual correctness (~74–75%) and a stable Inter-Annotator Agreement (IAA) of 63.92%, producing a dependable training/evaluation corpus for sarcasm classification models.
