Take your results with you

Goal: receive three consistent files and verify you can map every result to its original patient.

1. Choose where files go

Before processing, select a writable output directory. The CLI writes files on the machine where it runs, even when the API or inference server is remote. Keep /status watch active or resume it after reconnecting for automatic delivery.

2. Choose the intended values and layout

Automatic delivery defaults to Original · unreviewed. Use detailed findings for one row per finding, or document matrix for one row per document/run. For cohorts with completed reviews, you can choose reviewed results instead. Values still awaiting a decision remain marked as unresolved.

3. Verify completion

Look for successful CSV, Excel, and JSON downloads in the summary. Open the output directory and inspect the manifest. Check that patient_id contains the original identifiers, including 00123 in the fictional example. In Excel, use the XLSX file directly; when importing CSV, set ID columns to Text.

4. Recover an interrupted download

Enter /export for the selected cohort and choose the offered resume or verify/restore action. Choose the interrupted download so Ember restores the missing files. Rerunning models is not necessary to recover an export.

Files and provenance

Completed export delivery — Synthetic CLI demonstration · select image to view full size.

Automatic delivery writes a separate snapshot directory under the output folder you selected:

text

YOUR_OUTPUT/
└── COHORT_ID/
    └── DELIVERY_ID/
        ├── results.csv
        ├── results.xlsx
        ├── results.json
        ├── manifest.json
        └── snapshot.zip

CSV, Excel, and JSON come from the same extraction snapshot. The manifest records context, counts, and checksums; the ZIP supports recovery. Each download has its own folder, so earlier files stay intact.

Original · unreviewed is the default. Unsaved or unfinished reviews do not change these files.
CSV and Excel include patient_id first, with document/run/prompt identity and provenance. Failed documents remain visible.
Choose detailed findings or a document matrix. Excel stores IDs as text; when importing CSV into Excel, select Text for ID columns to retain leading zeros.
Previously reviewed cohorts still have explicit reviewed-export commands.