Publishing from here is a mutation — every action is confirmed and written to the FRELIP Agent OS audit ledger.

Give detailed instructions for revising a draft. The agent applies them in line with the FRELIP-NNFA house style (HTML-infographic, SEO/GEO/AEO, full metadata, verified official media + CTA, comprehensive, humanised, why-it-matters, no libel, DRAFTS-ONLY).

Every request is written to the FRELIP Agent OS audit ledger and the revision queue.

Author voiceprint provided (voice-first workflow) No voiceprint → prompt scrub ceilings ~31% human; with voiceprint → <10% AI target.

Optional external: Quillbot Grammar — ⚠️ text you paste there leaves this server. Use only for non-confidential snippets. Grammar Check above is fully local.

Governance: a detector score is never evidence of authorship. Everything here is a lead for a human to verify — not a finding, not a gate input, and never to be quoted to an author or reviewer. Detectors are unreliable under paraphrase and are biased against non-native and formal academic English (Liang et al. 2023: 61.22% average false-positive rate on 91 human-written TOEFL essays). Authorship questions are resolved by the Author Provenance Record.

Report appears here. Citations are leads requiring human verification — the desk never concludes fabrication on its own.

local = RAG (top-k evidence → cited answer) · global = GraphRAG (Leiden community summaries → synthesized answer). Answers are generated by the gateway LLM grounded ONLY in the retrieved evidence, with in-line source citations. Answers can take ~30–90s.

Ask a question to receive a grounded, source-cited answer.