Collinear Data Catalog
Production-grade task environments for RL capability development. Each domain includes simulated enterprise applications, realistic seed data, NPC behavioral simulations, and automated verifiers.
Coding
Simulate coding tasks from simple bug fixes to complex multi-feature implementations, using the industry-standard tools engineers rely on every day.
Memory (Multi-Episode)
Tasks that span multiple sessions or episodes, requiring the agent to retain and apply context across interactions.
Long-Horizon (>100 tools)
Multi-step workflows requiring 100+ sequential tool calls, where the agent must maintain coherent state and goal-tracking across an extended action chain.
Artifact Generation
Tasks requiring the agent to generate structured outputs: reports, spreadsheets, charts, and formatted documents ready for real-world use.
Error Recovery / Self-Correction
Reconciliation, debugging, and self-correction under failure states — tasks where the agent must detect what went wrong and fix it without human guidance.
Ambiguity Resolution
Classification and triage with incomplete or contradictory inputs, where the agent must make confident decisions despite missing information.
Planning & Re-planning
Coordination, roadmaps, and scheduling with dynamic constraints — tasks where plans must adapt as new stakeholder inputs arrive.
Tool Creation
Tasks where the agent must build or extend its own tools to complete objectives that go beyond its default capabilities.
Multi-turn
Tasks spanning multiple NPC personas and extended conversation turns, each requiring a different communication approach from the agent.