EviTra: Bridging the Cognitive Gap in Evidence Synthesis through Adaptive Intelligence

Abstract: The exponential growth of scientific literature has created a critical "evidence bottleneck," where the manual synthesis of research is increasingly prone to inquiry bias, reviewer fatigue, and systemic reproducibility failures. This whitepaper introduces EviTra (Evidence Tracer), a Systematic Review Automation (SRA) framework designed to bridge this cognitive gap through an adaptive, model-agnostic architecture. Central to the system is a Multi-Formalism Architecture that dynamically detects study archetypes and maps them to appropriate frameworks—moving beyond clinical PICO templates to encompass PECO, SPIDER, SPICE, and ECLIPSE formalisms. This ensures that qualitative nuance is preserved alongside quantitative data. To address the performance variance across the burgeoning AI landscape, EviTra implements a "Cognitive Tuner," which optimizes prompt logic through three distinct strategies: "Precision Mode" for atomic extraction with small edge models, "Balanced Mode," and "Reasoning Mode" for deep chain-of-thought analysis with large-scale engines. The framework modularizes the synthesis pipeline into the Study Summariser, which utilizes a recursive strategy to establish "Truth Anchors" and detect outcome reporting bias, and the Evidence Collector, which employs a "Data Clerk" persona and a zero-shot math prohibition protocol to ensure absolute numerical fidelity. Technical rigor is further maintained through a proprietary XY-Sorting algorithm for robust PDF reconstruction and a deterministic enforceLogic protocol that aligns automated assessments with GRADE certainty standards. By integrating human-in-the-loop verification via an interactive Evidence Matrix, EviTra transforms unstructured PDF repositories into rigorous, queryable databases, effectively turning a months-long manual slog into an accelerated, reproducible, and transparent evidence-to-decision pipeline. Access to repository and link to the web version of the software for testing, can be obtained by contacting the author.