# n8n synthetic failure baseline

Status: `VERIFIED`

This evidence package belongs to the first AI Brief Note reproducibility run.
The workflow uses deterministic, synthetic failures so that the recording
pipeline can be verified before testing a third-party API.

## Planned protocol

- Pin the n8n container image and record its digest.
- Execute the same workflow 100 times.
- Pass `EXPERIMENT_RUN=1..100` into each isolated execution.
- Inject a timeout classification on multiples of 25.
- Inject a 429 classification on other multiples of 20.
- Record exit code, wall time, error classification and a SHA-256 hash of the
  raw output.
- Publish the raw outputs only after checking them for secrets and local paths.

## Current files

- `workflow.json`: importable n8n workflow.
- `manifest.json`: protocol and target environment.
- `runs.csv`: generated only after the real run.
- `summary.json`: generated only after the real run.

## Result

- 100 total executions.
- 92 successful executions.
- 4 planned `INJECTED_HTTP_429` failures.
- 4 planned `INJECTED_TIMEOUT` failures.
- 0 unexpected failures.
- Workflow execution time: P50 934 ms, P95 1,097 ms.
- CLI wall time: P50 3,363.33 ms, P95 4,062.74 ms.

The 92% figure is not an n8n reliability rate: eight failures were deliberately
injected by the protocol. This baseline only verifies that the workflow engine,
CLI status and evidence collector distinguish the planned branches.

## Limitations

- The workflow uses deterministic synthetic failures, not a third-party API.
- Network access was disabled.
- The timing includes n8n CLI process startup inside one persistent container.
- Retry and recovery behavior were not tested.
- Environment-variable access was enabled only for the public synthetic run
  index. This is not a recommendation to expose production secrets to Code
  nodes.