Skip to content

README.md

Latest commit

 

History

History
87 lines (58 loc) · 1.71 KB

README.md

File metadata and controls

87 lines (58 loc) · 1.71 KB

Retry Loop Replay

Replay a stored retry-loop execution trace through X-Ray.

The fixture contains a provider-backed retry workflow trace captured from a real multi-step execution.

Replay

CLI replay:

python -m cli.main examples/retry_loops/retry_loop_trace.json

SDK replay:

python examples/retry_loops/retry_loop.py

Optional live-capture regeneration:

python examples/retry_loops/generate_retry_loop_trace.py

Optional fixture verification:

python examples/retry_loops/verify_retry_loop_example.py

Execution Pattern

The trace demonstrates a retry-collapse pattern commonly observed in multi-step LLM workflows:

  • retries continue
  • reformulations vary slightly
  • execution remains locally coherent
  • token usage increases
  • marginal contribution declines

This execution shape commonly appears in:

  • retry loops
  • recursive execution chains
  • repeated tool-call workflows
  • iterative refinement systems
  • long-running agent orchestration

Example replay verdict:

[VERDICT]
Execution should have stopped at Step 2.

[WASTE]
68% of execution happened after peak contribution.

[TIMELINE]
Step 1 → Improving
Step 2 → Peak
Step 3 → Repeating
Step 4 → Repeating
Step 5 → Declining

CLI Replay Output

CLI Replay

UI Replay Output

UI Replay

The local replay UI visualizes execution trajectories, contribution progression, redundancy growth, and peak-step transitions from deterministic replay traces.

Trace Artifacts

  • retry_loop_trace.json
  • retry_loop_live_raw.json

Related Examples

  • examples/langchain_callback/
  • examples/crewai_callback/
  • examples/iterative_refinement/