Model Adaptation · Data artifact
Agent Trajectory Dataset
Data artifactModel AdaptationModelsarc:AgentTrajectoryDataset
A training dataset of complete agent trajectories, sequential records of observations, reasoning chains, tool selections, actions and outcomes, including both successful and failure trajectories.
Responsibility. Supplies process-level supervision for agent fine-tuning.
Also known as: Trajectory data, Successful and failure trajectories, SFT training set, Expert demonstration dataset, DAgger aggregated dataset
Relationships
is read by dependency
is written by dependency
receives data from dynamic
is evaluated by assurance
is produced by lifecycle
Design guidance
- SHOULD capture the full reasoning and tool-use sequence rather than only input-output pairs.
- SHOULD span diverse task categories to teach generalizable reasoning patterns.
- SHOULD start with 100-1,000 trajectory pilots before scaling to production datasets.
Quantitative guidance
As stated by the sources; verify before use.
- AgentBank holds 50,000+ trajectories across 16 task categories; models fine-tuned on it gained 40-50% over base models on held-out tasks (Ch3.5).
- Fewer than ~1,000 trajectories often underperform few-shot prompting; gains plateau around 5,000-10,000 quality trajectories with logarithmic scaling (Ch3.5).
- Production fine-tuning datasets typically require 5,000-50,000+ trajectories (Ch3.5).
Classification
- Patterns
- Learning from agent trajectoriesNegative examples from failure trajectories
- Technologies
- AgentBank
- Quality attributes
- Functional suitability: correctness and validity (ISO/IEC 25010 | NIST AI RMF: valid)
- Risks mitigated
- Learning outputs without the reasoning processNarrow task-specific heuristics
Sources
- Ch3.5: T. Nguyen, "Prompt Optimization, Few-Shot Learning, Fine-Tuning," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 3.5. ISBN: 9798244538229.
- Ch5.12: T. Nguyen, "Learning-Based Decision Making Fundamentals," in Mastering Agentic AI Systems: Guide for the NVIDIA NCP-AAI Exam, 1st ed. 2026, ch. 5.12. ISBN: 9798244538229.