# Review delivered-data quality before a customer meeting


> Production does not yet enable the `submission:read` Key route in this baseline. Use a separate signed-in session for submission inspection; the delivered-record export still needs its Data Key.

> Before using these instructions, check [operation availability and compatibility](/api/compatibility).

When a customer flags records, start with the Campaign IDs, subscription scope, dates, and row IDs in their feedback. A data-read API Key can replay records **delivered** through its authorized subscriptions. It does not show every submission or another organization's records. Write the scope and reporting period in the report before calculating a rate.

For each Campaign, use a full organization API Key with Data access to the subscription, obtained through [Get your first authorized data](/api/api-data). In bash or zsh, enter that Key at the hidden prompt below; it is passed only to the named commands. The Key and IDs must belong to production, the API host shown here. Keep the Key in your approved secret store and the exported records in the private directory.

```sh
export HB_ENVIRONMENT='production'
export HB_API_URL='https://api.humanbased.ai'
unset HB_API_KEY
printf 'Data Key: '
read -rs HB_API_KEY
printf '\nCampaign ID: '
read -r HB_CAMPAIGN_ID
printf 'Subscription ID: '
read -r HB_SUBSCRIPTION_ID
HB_AUDIT_DIR="$(mktemp -d)"
printf 'Export directory: %s\n' "$HB_AUDIT_DIR"
hbd --version
HB_API_KEY="$HB_API_KEY" hbd auth whoami --json
HB_API_KEY="$HB_API_KEY" hbd campaign subscription list "$HB_CAMPAIGN_ID" --json
HB_API_KEY="$HB_API_KEY" hbd campaign records pull "$HB_CAMPAIGN_ID" --subscription "$HB_SUBSCRIPTION_ID" \
  --limit 200 --all --json > "$HB_AUDIT_DIR/campaign-records.json"
printf 'Exit code: %s\n' "$?"
```

[Export every delivered record](/api/cli-export) explains continuation and file handling. Record the exit code, and require `data.complete=true` before calling the file a complete read of that subscription's currently readable records. Repeat for every relevant Campaign and subscription; one export does not cover the whole organization.

If this Key also has `campaign:read`, establish a Campaign-level baseline with the same half-open event-time window:

```sh
HB_API_KEY="$HB_API_KEY" hbd campaign metrics "$HB_CAMPAIGN_ID" --from <start-iso> --to <end-iso> --json
HB_API_KEY="$HB_API_KEY" hbd campaign outcomes "$HB_CAMPAIGN_ID" --from <start-iso> --to <end-iso> --json
```

Keep each command's denominator and time basis. These counts describe Campaign events; the delivered-record export describes one subscription's readable rows. Do not divide one by the other as though they were the same cohort.

Make a row-level review sheet with `campaign_id`, `subscription_id`, `submission_id`, the date and field being checked, and the evidence reference. Compare the delivered fields with the agreed output schema and the customer's row IDs. Check required fields, date windows, repeated transaction identifiers and normalized addresses, and exchange or network labels. Mark these as **review candidates**. A repeated address may be permitted work, and a quality grade or successful pull does not prove customer ingestion or real-world wallet ownership.

A person with a role on the Campaign can inspect submitted evidence and the review outcome in a **separate signed-in CLI session**. An organization API Key can do the same with the `submission:read` grant, limited to its organization and campaign allowlist:

```sh
hbd campaign submissions list <campaign-id> --page 1 --limit 100 --json
hbd campaign submissions inspect <campaign-id> <submission-id> --json
```

Run them with `HB_API_KEY` absent to select the signed-in session, or provide an organization API Key with `submission:read`. For guarded or sealed Campaigns, the signed-in person may add `--reveal` at the terminal to view hidden answers; a Key always receives those fields hidden. Sample flagged and unflagged rows across Campaigns and contributors. Reconcile the delivered IDs with the customer's ingestion log, record the reviewer's decision and any unresolved customer policy, and only then state a quality conclusion. Reading and reporting do not change verdicts or rewards. Production has not yet deployed this Key grant; until rollout, use the signed-in session there.
