From Search to Synthesis: A Living Research Platform for AI-Augmented Evidence Mapping

Systematic reviews remain the gold standard of evidence synthesis, yet they are increasingly mismatched to the scale and pace of contemporary research. A typical review takes 12-18 months, produces a static snapshot, and cannot feasibly screen the tens of thousands of records now common in interdisciplinary fields. The recent literature shows a field converging, independently and across disciplines, on a set of partial answers: retrieval-augmented generation that grounds machine-generated synthesis in verifiable sources, multi-model consensus schemes that keep a human in control, and automated trend-and-gap detection over large corpora. These advances have so far emerged as disconnected components without shared validation infrastructure, reproducibility protocols, or mechanisms for continuous updating. Living systematic reviews offer a conceptual solution to the currency problem but lack the computational infrastructure to execute at scale. We present LivingMeta, an open research platform that integrates the methodological pillars the field is converging on — structured multi-persona extraction, automated evidence gap analysis, adaptive evidence communication, guided gap-closing research, and instance-owned publication — into a single reproducible workflow. The platform implements triangulated AI extraction through six cognitively differentiated personas with field-level consensus scoring, a model-adaptive transparency score (LMQS), an eight-type gap taxonomy that extends prior frameworks, an optional bring-your-own-data (BYOD) corpus layer, and a collaborative architecture in which human researchers and AI agents iterate through a state-machine-governed research lifecycle with role-adaptive coaching. The architecture currently indexes over 500,000 classified papers across seventeen deployed instances spanning innovation management, sports analytics, AI education, public governance, art history, research integrity, responsible AI, the literature on AI in research itself, and additional domains, with a shared extraction store enabling cross-domain synthesis. This paper is foundational rather than evaluative: it specifies and justifies the platform's design decisions against the convergent recent literature, and reserves systematic accuracy benchmarking for future work. We argue that a research synthesis platform constitutes a methodological contribution when it structures research through explicit protocols, enforces quality through architectural constraints, and enables reproducibility through logged, versioned workflows.