Upload documents (PDF, DOCX, TXT, MD), structured datasets (CSV, JSON/JSONL), code, configs, or LaTeX. Deep Synthesis will chunk, cross-reference, and synthesize the session sources into provenance-backed hypotheses, contradictions, null results, and next-step validation plans. When citation graphs or other network structure are present, hub-compression analysis runs automatically to surface high-signal clusters instead of repeated evidence. For datasets, name the target or outcome column in your prompt when you can (for example, PSP_status, tau_burden, or cell_viability). High-confidence hypotheses can then route into Research Lab validation for matched model tests or symbolic-regression checks. Code-backed hypotheses can route into replay or ablation-style validation. You can also discuss any finding with the Companion to turn the result into durable working memory.