Import
Use the incident JSON format or an existing RunLedger JSONL run.
PRODUCTION INCIDENT IMPORTS
Your next regression test
already happened.
Bring a production agent trace into RunLedger. Redact it locally, replay the recorded tool responses in order, and keep the failure as a test your team can run again.
Your production logs stay on your machine. This page does not upload or process them.
Redact & review
Replace sensitive values while preserving their relationships in the trace.
Reproduce
Run your agent against the recorded responses, one occurrence at a time.
Keep the check
Promote the incident and gate CI on the business rules you defined.
FOLLOW THE EVIDENCE
A familiar failure.
A smaller, reviewable bundle.
This production-format refund trace uses invented customer data. The downloadable CLI imported and redacted it, then replayed its agent. The browser displays those saved results; it does not execute the replay.
Loading the imported example…
The candidate follows the recorded calls and final output.
A reproduced failure is still a failing regression check.
Effects come from the supplied ledger, not inferred tool success.
WHAT LEAVES THE IMPORT STEP
Same relationships. Replaced values.
BEFORE · INVENTED EXAMPLE DATA
AFTER · LOCAL REDACTION
Repeated identifiers keep a consistent replacement within the bundle. Known credentials and personal fields are redacted; free-form text is removed by default. Redaction can change behavior that depends on those values.
Inspect the redaction report
ORDERED TOOL REPLAY
The first timeout stays first.
Repeated calls consume separate recorded responses. An unrecorded or changed call stops replay instead of reaching a live tool.
Inspect the sanitized events
CHECK THE BUSINESS RULE
Reproduced does not mean resolved.
The original failure remains visible. Use --gate to fail CI when the replay diverges or its business rules still fail.
Inspect the recorded effect ledger
MAKE THE FAILURE REPEATABLE
Import locally.
Review before sharing.
Install or upgrade RunLedger from PyPI, then follow the guide to map your trace to the supported incident format and connect your JSONL adapter. Use an active Python environment.
RunLedger on PyPI Read the production incident guideRequires RunLedger 0.3.0 or later. Existing pipx users: pipx upgrade runledger. The optional source and example download includes example files that are not installed by pip.
# Install or upgrade the CLI
python -m pip install --upgrade runledger# Explore the incident commands
runledger incident --help# Import and redact a local trace
runledger incident import incident.json --out sanitized# Inspect the sanitized evidence
runledger incident inspect sanitized/incident.json --out inspection# Preserve it for future replay
runledger incident promote sanitized/incident.json --to regression.jsonA CLEAR REPLAY BOUNDARY
Reproduce what was recorded.
Review the redaction
Automatic redaction is a starting point, not a guarantee that every sensitive value is removed. Add paths for your schema and inspect the exported bundle before sharing it.
Keep the missing state visible
Replay freezes recorded tool responses. Model calls outside the adapter, missing effects, and unrecorded external state are not reconstructed. A different branch is reported as a divergence.
Verify fixes with the right test
An incident replay checks fidelity to the captured behavior. Use the stateful support-agent scenarios to exercise a changed policy against a modeled world.