Make AI research agents accountable — give every conclusion a traceable argument graph. MCP server + Claude Code plugin for paper reproduction, hypothesis verification, and auditable scientific reasoning.
Use when drafting or revising manuscript text that must project graph-backed scientific arguments into paper prose across Introduction, Related Work, Results, Discussion, Methods, figures, tables, source data, and citation handles.
Use for open-ended autonomous research where you must generate the research questions and Claims yourself from provided data and reference material. Route the goal through a Framing step, propose falsifiable Claims, produce observed Grounds from the data and literature Grounds from the references, actively seek…
Use when citations in a LaTeX manuscript need checking against what the cited papers actually say — prose that may overstate, understate, or mischaracterise a source. Not for writing new prose or creating citations; that is literature-writing.
Use when surveying literature at scale — dozens to hundreds of papers — to build a taxonomy, locate a gap, establish a trend, or screen a large candidate pool. Not for prose; hand off to literature-writing for the write-up.
Use when writing or revising LaTeX prose that cites published papers — related work, introductions, discussions, or the write-up phase of a survey. Translate citable propositions into verified literature Grounds and cite them as \cite{statementN}.
Use once when a LaTeX project needs Overleaf synchronization with Warranted citation enforcement. Installs/authenticates leaf, binds a local LaTeX directory to an Overleaf project, and writes the Stop hook that pushes staged source with statementN citations expanded.
Use when reproducing a published paper, or judging whether its Claims hold under independent execution — including a run that disagrees with the paper, or one that appears blocked. Not for testing a hypothesis of your own.