AI & Agentic Update Policy

We use AI, transparently

Our report engine uses large language models — primarily Claude by Anthropic — to accelerate research synthesis and draft structured analysis. AI is a tool, not the author of record.

What AI does

• Synthesises primary and secondary sources into a structured first draft.
• Applies our internal prompt taxonomy (categories, sections, framing rules).
• Checks reference hygiene against configured validators.

What AI never does

• Auto-publish. Nothing leaves the platform without human sign-off.
• Fabricate statistics, cases, or citations. Where we detect fabrication risk, we block publication and regenerate.
• Replace the analyst's judgement on what to cut, what to soften, and what to sharpen.

Model changes

We update models and prompt taxonomies on a rolling cycle. We maintain an internal log of model/version per section so that if an issue is discovered with a particular model generation we can identify and republish the affected reports. Buyers of affected reports are notified.

Agentic features

Some parts of the engine use autonomous agents — research queues, safety-net pollers, refresh jobs. Each agent has a defined, narrow role, runs on an auditable schedule, and cannot publish content without human review.