# Celery queue-routing lab, November 13, 2026

Synthetic reproduction, not a production installer or customer incident. Two baseline and two mismatch/repair phases passed on Linux with Python 3.12.16, Celery 5.7.0, redis-py 8.1.0 and Redis 7.4.12. No PingCron endpoint, real alert or signup is exercised.

## Evidence inventory

- `evidence/report.json`: aggregate result, exact commands, IDs, timestamps, versions and tested source hashes.
- `evidence/runtime.json`: runtime metadata.
- `evidence/installed-requirements.txt`: dependency versions installed in the observed run.
- `evidence/1-baseline`, `1-wrong_queue`, `2-baseline`, `2-wrong_queue`: each contains `beat.log`, `worker.log` and `report.json`.
- `source/app.py`, `run.py`, `test_safety.py`, `requirements.txt`: original tested fixture code. Hashes for app.py/run.py/requirements.txt match the result. Schedule database files are omitted because they are generated state, not reproduction inputs.

Provenance: Tnellyy/pingcron commit `345fd555b7917c7ff063815a2ded93146c855890`, Actions run `43475969961`, job `132059745033`. The repository is private, so the public bundle supplies inspectable source and traces instead of requiring repository access. Original CI artifact SHA256: `0b1b8548de9a53bc51eafa48b52bb80e4bbe1dca4bddfdb8808dd145771b0a0f`. This public bundle differs because it adds source and omits generated schedule files.

## Isolated reproduction

Use a fresh directory in a disposable Linux environment with Docker and Python 3.12. Do not use a shared broker, even if it is forwarded to localhost. Unset inherited CELERY_* settings. The fixture refuses them and requires explicit acknowledgement. The Redis container below has a dedicated name and a loopback-only port. If that name or port is already occupied, stop and choose an unused name/port consistently; do not remove someone else's container.

```sh
docker run --detach --rm --name pingcron-celery-lab-20261113 \
  --publish 127.0.0.1:16379:6379 \
  redis@sha256:7f7d30b7fdd09467bac58198a1f166dd69789d348d4fe859cccb20e0cf287fbe
docker exec pingcron-celery-lab-20261113 redis-cli ping
python3 -m venv .venv
. .venv/bin/activate
python -m pip install -r evidence/installed-requirements.txt
export LAB_REDIS_URL=redis://127.0.0.1:16379/0
python source/test_safety.py
python source/run.py --ack-disposable-broker
```

Wait for the Redis check to return PONG before running the fixture. The digest above is the image pulled by the recorded CI run. The Python lock records observed versions; package availability and platform still affect future installation. This is not a byte-identical reproducibility guarantee.

The harness creates `celery-routing-evidence`, refuses to overwrite it, and runs each phase with unique names and bounded deadlines. It stops its own worker and Beat process group. After collecting results, remove only the disposable container you created:

```sh
docker stop pingcron-celery-lab-20261113
```

No Redis flush, queue purge or external workload is used. A failed run must be investigated from its own traces, not counted as a passing repeat.

## Interpretation

Both fault phases show the producer's original IDs queued while the intended worker consumes another queue and both targeted probes answer. After Beat stops, adding the intended consumer lets those same IDs start and complete with sum50 and row_count3. Baselines prove the configured task can run. This isolates a subscription mismatch in this fixture.

The experiment does not test arbitrary broker failures, Django database schedulers, durability, exactly-once processing, retries, expiry, output correctness beyond this fixture, performance or PingCron alert delivery. The process leaders report stopped; cleanup also targets their process group, but this run did not independently census descendants or inject a timeout failure.
