Abstract: Peer review and evidence synthesis frequently encounter the Missing PECO problem, where research objectives are implicit or fragmented across manuscript sections. Manual reconstruction of these frameworks often introduces Inquiry Bias, inadvertently filling in protocol details based on reviewer expertise rather than author text, which undermines the reproducibility of scientific audits. This concept paper introduces EviTra (Evidence Tracer), an architectural framework designed to automate the extraction and cross-sectional comparison of evidence elements. This proof-of-concept demonstrates an architectural pivot from passive auditing to an end-to-end Systematic Review Automation (SRA) pipeline. The engine utilizes a proprietary XY-Sorting algorithm for robust PDF reconstruction and a Scorched Earth linguistic policy to filter statistical noise. We introduce two specialized downstream modules: the Study Summariser, which employs a recursive extraction strategy and a Structured Abstraction Protocol to detect spin through an automated alignment check; and the Evidence Collector, which utilizes a Data Clerk persona with zero-shot math prohibition to harvest quantitative effect estimates with absolute numerical fidelity. Data integrity is maintained via a deterministic enforceLogic protocol that aligns automated ratings (Levels 1–4) with standardized GRADE certainty actions. System transparency is achieved through a Semantic Pipeline Trace and an interactive, contenteditable Evidence Matrix for human verification. The architecture establishes scientific rigor via local server deployments (LM Studio) and high-level reasoning through Google Gemini, providing a scalable solution to the cognitive bottleneck in large-scale toxicological reviews. EviTra transforms from a passive auditing tool to an active participant in the data extraction lifecycle, effectively decoupling qualitative synthesis from the rigid precision required for meta-analysis.Access to repository and link to the web version of the software for testing, can be obtained by contacting the author.
Building similarity graph...
Analyzing shared references across papers
Loading...
Jan-Peter Nilsson (Sun,) studied this question.
www.synapsesocial.com/papers/6994055d4e9c9e835dfd6372 — DOI: https://doi.org/10.5281/zenodo.18633185
Jan-Peter Nilsson
Rio Tinto (Australia)
Building similarity graph...
Analyzing shared references across papers
Loading...