# Reproduce the memory and MCP checks

These checks use fictional sources in temporary vaults. They test local components and real MCP SDK calls. They do not run a language-model agent conversation.

The recorded run used the Full 1.1.0 archive, Python 3.11.14 and MCP SDK 2.1.1.
Its execution date is 2026-09-14. The result identifies the archive and test hashes.

1. Download the [operations tools bundle](https://gigabrain.download/assets/resources/editorial-operations-tools.zip).
2. Extract it into a separate directory.
3. Open a terminal in that directory.
4. Create a Python 3.11 virtual environment.
5. Install the recorded MCP dependency in that environment.
6. Choose an output file that does not exist.
7. Run the command with your matching customer archive.

```sh
python3.11 -m venv .venv
.venv/bin/python -m pip install mcp==2.1.1
.venv/bin/python dev/public_editorial_operations.py \
  --archive /path/to/gigabrain-full-1.1.0.zip \
  --output /path/to/new-operations-result.json
```

The runner extracts the archive into temporary storage. It copies the supplied tests beside that product copy.
The tests create fictional sources, approved synthetic changes and isolated project scopes.

Review the individual test names and output. Expected permission failures appear in the log.
A skipped protocol test does not count as a passing run.

The deletion-plan test checks that the source stays present. The removal check covers retired graph support.
Neither check proves removal from Git history, backups, external originals or provider systems.

[Inspect the recorded result](https://gigabrain.download/assets/evidence/editorial-operations.json).
Use its actual archive identity when reproducing an older result.