Curated multi-environment scenario packs with prompts, verifiers, seeds, golden traces, and run evidence. Buy only the layer you need, from prompt-only tasks to full Harbor environments.
42 tool calls in golden trace. SQL + rubric verifier. Seeded actor persona. Optional environment.
Start with prompts, add verifiers when you need measurable evaluation, or license full executable packages when your team wants reproducible agent runs.
Fifty executable SpotHub computer-use tasks with the shared SpotHub desktop runtime and per-task run evidence.
Fourteen long-horizon workplace tasks spanning email, project, calendar, and chat surfaces on the shared four-gym runtime suite.
Three multi-surface Default-DB evaluation tasks packaged with shared runtime gyms and the Default-DB seed dataset bundle.
Each pack is designed to support training and evaluation decisions: realistic workflow context, objective checks where possible, and evidence that the task stresses current agents.
Human-reviewed scenario design, not raw synthetic dumps.
Model-attempt evidence can be attached before purchase.
Buyer can choose prompts only, prompts plus verifiers, or full executable task package.
Harbor-compatible packaging keeps the data useful for RL, SFT, and eval workflows.
Select from operational domains where long-horizon tool use matters: coordinating people, updating systems, reconciling context, and leaving an auditable final state.
Tell us what your team is evaluating or training. We will use this to recommend an existing pack or scope a custom long-horizon workflow dataset.