Using

Exports and replay

Download the test pack and replay it locally.

On this page

Download the tests
Rerun them on your machine
Understand the command result
Check the file you received

Download the tests

A test pack contains the saved fictional requests, approved checks and references to the original evidence. It does not include model keys, raw Qloo responses, the assistant’s answer text or executable code. Treat it as private project data.

Rerun them on your machine

Replay means running those downloaded tests again. You need your local model-provider settings, but no TasteCase login or runner token. Replay uses the saved inputs and does not ask Qloo for fresh evidence.

Replace the file paths and quoted checksum placeholder

export TASTECASE_ASSISTANT_VERSION="concierge-prompt-v2"
node packages/runner/dist/cli.js replay \
  --pack ./tests.json --role candidate \
  --pack-checksum '<trusted pack checksum>' --report ./new-report.json

This command runs only Candidate. Baseline stays pending, so the report is partial. Use a new report filename each time; existing files are not overwritten.

Understand the command result

Code	Meaning
0	The selected version completed without a rule failure. Advisory findings still need review.
2	An approved rule failed.
4	A run was unavailable or did not finish.
1	Check the input or configuration; the command may also have been cancelled.

Check the file you received

Keep the checksum shown after your signed-in download and use it with replay. It helps detect a changed file; a checksum stored inside the file alone is not proof of who created it.

Note

“Pack downloaded” means you received the file, not that you ran it. TasteCase cannot see a local replay unless you separately share its report.

For exact file formats, see Artifact schema.