Adaptive Research Paper Collaboration System

Problem

Research data analysis and paper writing often depend on judgment: what file format is present, whether the data are usable, which statistical or predictive methods fit, whether figures match the evidence, and whether the conclusions are internally consistent. Hard-coding file types, expert roles, chart types, or paper categories makes the system brittle on new tasks. paper-agent replaces fixed presets with an evidence-first reasoning workflow.

How it works

• Zero-preset file intake: it inspects file headers, NULL bytes, delimiter density, candidate encodings, and read-back structure to decide whether data are CSV, TSV, JSON, binary, or another format; after three failed attempts it asks for help.
• Deterministic profiling: it computes row counts, missingness, column types, numeric summaries, unique values, normality signals, outliers, and correlations, reporting only what is observed.
• Task inference and role generation: it derives likely research goals, data structure, and method needs from the user's description, then generates roles for statistics, domain knowledge, or visualization as required.
• Three-layer validation: it checks numeric consistency, verifies assumptions, cross-validates with alternative methods, and requires each conclusion to trace back to evidence.

Boundaries

It is useful for research-data triage, analysis planning, paper consistency checks, and evidence tracing. It does not replace professional medical, legal, or domain review. When the file is unreadable, the task is ambiguous, or key assumptions cannot be verified, the skill should surface uncertainty and request human input.