research

# Research Corpus Analyzer Structured multi-step analysis system for academic and research corpora. Separates fact / inference / unknown at every step to prevent hallucination. ## Universal Prefix Prepend this to EVERY prompt in the chain: ``` Work only from the uploaded materials. Explicitly tag every claim: - <fact> — directly confirmed in the corpus - <inference> — reasonable conclusion from the corpus - <unknown> — insufficient data to conclude Rules: 1. Do not invent sources, authors, quotes, consensus, or gaps not in the corpus. 2. For each strong claim, state which works support it. 3. If the corpus cannot support a confident conclusion, say so explicitly. 4. Do not use external knowledge unless separately requested. 5. Do not summarise in generalities where you can show structure, contradiction, or limitation. ``` --- ## Modes ### /research quick Runs steps 1 → 6 → 10 only. Use when: fast overview needed, corpus is small (< 5 papers), time-constrained. ### /research full Runs the complete chain: **1 → 2 → 5 → 7 → 4 → 6 → 8 → 3 → 9 → 10** Use when: serious research work, thesis support, deep literature review. ### /research mode:[N] Runs a single step N (1–10) in isolation. Use when: user already has prior steps done and needs one specific analysis. ### /research stop-check Runs step 9 only. Use BEFORE committing to a full chain. If corpus coverage < 60% or critical bias found → stop and report what to add. --- ## Step Definitions ### STEP 1 — Corpus Inta...