# Reproduce the connected-source correction

This component test uses fictional project files and licensed Gigabrain archives. The bundle contains test scripts, not the paid product.

The recorded result ran on 2026-09-14 against Full 1.1.0 and System 1.1.0. It quotes source passages rather than generating an agent answer.

1. Download the [connected-source tools](https://gigabrain.download/assets/resources/connected-source-tools.zip).
2. Extract the tools into a separate directory.
3. Prepare the local encoder in an isolated Gigabrain installation using the archive's model setup instructions.
4. Open a terminal in the extracted tools directory.
5. Replace the two archive paths below with your matching licensed archives.
6. Replace the model-seed path with the prepared installation.
7. Choose an output file that does not exist.
8. Run the command below.

```sh
python3.11 dev/public_connected_reproduction.py \
  --full /path/to/gigabrain-full-1.1.0.zip \
  --system /path/to/gigabrain-system-1.1.0.zip \
  --model-seed /path/to/prepared-test-vault \
  --output /path/to/new-connected-result.json
```

The runner extracts each archive separately. It creates the fictional project, review and schedule, then indexes them.
The test retrieves the schedule through source-supported graph links. It changes 18 October to 25 October and checks both citations.

Inspect the original quotation, failed old receipt and amended quotation. Read the graph path's source support.
Graph links remain authored interpretations. A verified quotation does not establish that an inference is correct.

The new output includes the source files after correction. The original schedule appears in the saved original receipt.
The output's execution date belongs to your run. It does not replace the published result's original date or establish an agent-session result.

[Inspect the recorded source-correction result](https://gigabrain.download/assets/evidence/editorial-outcome.json).