The Editor Agent — From Draft to Publication-Ready
How Datarim's Editor agent brings written content to publication quality: fact verification with Chain of Verification, AI pattern removal, style consistency, and a structured editorial report.
AI-generated text has a recognizable shape. Certain sentence structures, certain vocabulary choices, certain formatting habits appear across models and contexts. Readers notice even when they cannot name it. The Editor agent exists to close the gap between a draft and something that reads as genuine.
Its work is editorial, not generative. The author's voice is treated as fixed. The agent corrects patterns, not style — it does not rewrite paragraphs or add new arguments. What it finds goes into a structured report; meaning-altering changes are flagged for author approval rather than applied silently.
Fact verification
The verification method is Chain of Verification: extract the verifiable claims, check each one against an authoritative source, assign a verdict. This is not a spell-check pass. A claim like "framework X has been adopted by 80% of Fortune 500 companies" might read plausibly but check false. The agent finds those claims before publication does.
The same method applies to cross-references and links. A link that resolves is not the same as a link that points to what the text claims it points to.
AI pattern removal
The audit runs in categories: vocabulary (banned filler words and constructions), structure (over-formatted lists, redundant section headers), formatting artifacts (unnecessary bold, excessive bullet nesting), communication tells (hedging phrases, performative enthusiasm), and linguistic patterns (passive-heavy sentences, nominalizations where verbs would be clearer).
Each category produces a count and specific examples. The report is structured so a human editor can scan it in two minutes and know exactly what changed and why.
Where it sits
The Editor agent runs at /dr-edit and participates in consilium for content decisions. It works natively in both English and Russian — language-specific AI patterns differ between the two, and the audit accounts for that. Before content moves to publishing, a Telegram-aware pre-publish review confirms length limits and checks for structural issues that would break platform rendering.
See the full reference on the Editor agent card, or read what Datarim is for context on how agents fit together.