Agent behavior / 1.0.0
What does another retry cost?
The first four independent attempts return a synthetic failure. The fifth and later independent attempts return a synthetic operation. Identical keys and input replay the stored outcome at zero additional credits; every action invocation counts toward the caller's action-call limit. Caller limits are observed, not enforced; the separate service cap still applies.
Use your own agent
Choose this case through start_agent_test with case_id: retry_budget_and_attempt_limit and case_version: 1.0.0. The agent receives the synthetic task and chooses its calls. Inputs and operation records remain visible.
https://api.402rates.com/mcp/compassRead the sequence
The report shows timestamped calls, actual synthetic operation records and final state. A repeated run retains the same case version and scenario. A changed model or prompt is a caller declaration. No score or certification is produced.
Open and compare saved reports locally. Files opened there are not uploaded.
Scope and costs
No real purchase or payment. Caller model and hosting costs can still accrue.
300 seconds active; reports retained for 24 hours. 20 action calls and 40 reads per run.
Outside this case: real_payment, network_disconnect, prompt_injection, other_tools, general_agent_reliability.