Help and analysis in context
Open Assistant or Help from the header, or Ask AI beside a workspace. Guide uses the current capability supplement, feature methods and dated handbook. Analyst works with approved public selections and source snapshots. Read citations, preview evidence and choose any proposed navigation or filter action explicitly.
Specialist evidence workflows
The suite includes score explanations, AI-assisted query expansion and semantic reranking, multilingual claim extraction, country-role review, report-linking candidates, contradiction/revision review, cited briefing drafts, alert triage and unsaved rule drafts, document excerpts, geographic queries and forecast-evidence review. Source enrichment remains advisory; original scores, records and forecast outcomes retain their existing definitions.
Document analysis supports bounded text and PDF excerpts with page references. Image-only excerpts use labelled AI transcription and need comparison with the original. Geographic results use deterministic calculations over eligible source-provided points; incomplete coverage stays visible.
Grounding and existing access
Answers carry source IDs, original supporting spans, dates, qualifications, model information and usage. Invalid references, unsupported numbers and support-review failures can withhold claims. These checks are not proof of publisher truth or calibrated predictive accuracy.
AI uses the signed-in account’s existing OpenAI connection and shared Oracle request/concurrency controls. Provider charges apply. Static guide content remains available without a model connection. This feature does not automatically read private portfolios, cases or browser storage, and it does not save or enable alert rules.
Start exploring
Refresh Watchtower, open Assistant, choose a tool and review the current context. Use Preview evidence before Run AI. Expand citations and inspect Plan & coverage for bounded selections, source gaps and document limits. Dedicated Oracle, Deep Analyst, native Data Q&A and existing saved workflows remain available.
Verification and availability
Desktop, tablet and mobile interface flows, source retrieval, access controls and bounded-response contracts were qualified. Model responses in software qualification were explicitly controlled fixtures; customer credentials and paid model calls were not used. Live generation depends on the signed-in account’s valid OpenAI connection and provider availability. These checks do not establish predictive accuracy or a zero-error guarantee.