Recalls. Federal contract recompetes. Biotech milestones. Acquisitions. Cassandra reads the leading signals buried in public data and surfaces what's coming — weeks early. One product, four verticals, and a single agent-callable endpoint.
Each vertical is its own live engine with its own data moat. They share a thesis, a track record, and a single MCP.
Consumer products, vehicles, drugs and medical devices showing sharply rising complaint velocity — a leading indicator of recall risk, weeks before any official CPSC / FDA / NHTSA action.
Every biotech catalyst starts years upstream as an NIH grant, then a trial registration, then a phase advance. We join those three public databases on a time axis and surface the upstream event the moment it lands — with the predicted downstream filing and a typical lead time.
Expiring federal contracts plus incumbent-performance and capture signals — flagging which recompetes the incumbent is likely to lose and who is positioned to take them, months before the solicitation drops.
Mid-tier government contractors showing the leading pattern of an acquisition target — backlog growth, set-aside graduation, PE ownership age and clearance density — before the deal is announced.
The research bench behind the predictions — discovers and back-tests the leading-signal correlations (which upstream public-data event reliably precedes which downstream event, and by how long) that power the other slices.
Integrate once. Call predict(domain, entity) and Cassandra routes to the right engine.
POST https://predict.dropwatchhq.com/mcp # JSON-RPC 2.0 (MCP)
tool: predict(domain, entity)
domain ∈ recall | bio | recompete | ma
→ predict("recall","Tesla Model Y") recall-risk signal + lead time
→ predict("bio","Moderna") NIH→trial→SEC catalyst chain
→ predict("recompete","") recompete-winner shift
→ predict("ma","") acquisition-target signal Every flagged entity is cross-checked against the official record. When it resolves, it becomes a permanent receipt: flagged on X, happened on Y, N days early.