Start with the investigation
Research tools now opens with Start research: investigate a country, track a topic/company, or review a forecast. Apply explicit choices, then follow links into the existing workspaces. Successful handoffs leave the destination visible; failures return a clear message rather than claiming completion.
Finance reference years and future target years are separate. A news topic is not silently used as a fiscal-series search. Standalone Futures steps use baseline assumptions and clear an older comparison partner. Dossier and watch steps prepare the existing controls; saving or enabling remains explicit.
Consistent applied source context
Research tools and Ask AI share context adapters for supported selections across Futures, geopolitical outlook and shortages, monetary/public-finance views, politics and government updates, country-profile overview, Pro Brief news preferences, Global Risk/Systemic, and public ratings/CDS references.
Local-only or unsupported selections remain visible as gaps. Pro Brief’s adapter shares its native news candidate query, which can span 72 hours; the brief separately applies its 48-hour window and display count. It is not an exact copy of the visible cards.
Map actions with checked outcomes
The existing Geopolitical Outlook, Global Politics and Monetary maps offer country-reference camera focus, a world camera view and restoration of the previous camera. The resulting camera state is checked before success is reported, with visible feedback on the map.
Assistant-proposed map actions require an explicit click. Owner/replay changes and newer interactions cancel stale actions; restoration is refused after a newer map gesture. Camera actions do not change source filters, scores or underlying observations. Country reference points are not incident locations.
Tested as connected tasks
Qualification exercises country and fiscal handoffs, separate forecast years, actual map controls, Assistant action dispatch, news-search preparation, provenance-preserving dossier saves and replay cancellation across desktop, tablet and mobile layouts.
Task timings measure these browser/source workflows on the test host, not human completion time, a service-level guarantee or a model-answer-quality benchmark. Private identities and stores were isolated; the AI action envelope was controlled, with no real paid model calls.