Retention Policy
Last updated:
Nugget is the service operated by the 25madison team. This policy explains retention of document-processing data, operational records, and recovery copies, and how to request scoped deletion. It complements the Privacy Policy and Data Processing Addendum. For support or deletion requests, contact connor.smith@25madison.com.
1. Scope and how to read the schedule
The schedules below describe defaults for Nugget's hosted service. Retention for a particular environment depends on enabled services, its configuration, and the applicable customer agreement. Customer-managed environments and copies exported to customers or their recipients are governed by their own settings and responsibilities.
Eligibility is not completed deletion. Google Cloud Storage (GCS) lifecycle rules use object age, rather than time since account closure or completion of a rights request. An object becomes eligible at the stated age; deletion occurs asynchronously afterward. The periods are not an SLA or a promise that every copy will be gone at that time. A workflow may create multiple objects with different creation times.
Ephemeral processing and temporary working data may be cleaned up earlier. These schedules do not require retaining each object for the full period, and ephemeral processing does not imply that usage records, logs, or provider-held copies are absent.
2. Hosted document and working objects
| Object category | Eligible at age |
|---|---|
| Job objects for extraction, OCR, and location workflows | 14 days |
| Uploads | 7 days |
| Caches | 30 days |
| Raw Document AI artifacts | 7 days |
| Scratch and classification working objects | 2 days |
| Idempotency objects | 1 day |
The raw Document AI entry covers artifacts stored in Nugget's GCS storage; it does not establish the Document AI provider's own retention. Cache eligibility likewise applies to the stored object category, not a universal retention promise for every in-memory or provider cache.
3. Usage, account, and operational records
| Record category | Handling |
|---|---|
| Firestore usage and quota ledger | 450-day time-to-live (TTL). TTL deletion is asynchronous, not a guaranteed completion time. |
| Daily, monthly, and account aggregate records | No automatic expiry. Continued retention or deletion is assessed on account closure or request. |
| BigQuery usage records, when enabled | 400-day default retention; the applicable environment configuration governs. |
| Application, infrastructure, security, and provider logs | Retention varies by deployment, log category, and provider. Contact support for the periods applicable to your service configuration. |
| Durable purge and deletion-control metadata | No automatic expiry. Restricted access; no document bodies by design. Continued retention is assessed against deletion tracking and control needs. |
Aggregates can remain linked to a tenant or account and are not necessarily anonymous. These records support usage accounting, quota administration, and service operations. Records without automatic expiry are assessed on account closure or request to determine what remains necessary and what can be removed.
Account and authentication metadata, credential records, and support correspondence are assessed for continued retention according to their purpose and the scope of an account-closure or privacy request. Revoking a credential, deleting document content, and deleting an authentication account are separate operations.
4. Backups and recovery copies
Hosted Firestore backup settings depend on the deployment configuration. Where enabled, the default backup schedule is weekly with 14-week retention. Point-in-time recovery (PITR) history follows the recovery window configured for the applicable environment. Contact support for environment-specific recovery periods.
The GCS soft-delete default is zero days for newly deleted objects. Soft-delete behavior follows the applicable bucket configuration. Copies created under earlier soft-delete or recovery settings expire separately under the settings that apply to them.
Removing live data does not necessarily remove backup or recovery copies at the same time. Their expiry follows the applicable recovery schedule. When previously deleted data is restored for recovery, applicable deletion instructions must be reapplied using deletion-control records.
5. Requesting deletion or account closure
Email connor.smith@25madison.com to request a purge, discuss retention, or close an account. Include your tenant or account context and the jobs, uploads, or categories at issue, where known. Do not include API keys, passwords, or document bodies. We verify your identity and authority, clarify the requested scope, and coordinate with the customer controlling the information where necessary.
The purge process is an offline, operator-assisted action scoped to identified live GCS and Firestore data. It is not an automatic, all-system erasure endpoint. The operator assesses the target scope, relevant retention needs, and available deletion results, and communicates what was addressed and what remains outside that scope.
Deleting an individual job is narrower than an account purge. The job-deletion API removes terminal job-owned records, inputs, results, and locks. It does not remove reusable uploads, shared OCR artifacts, or content-addressed caches. Active jobs or writers can prevent deletion, and a callback already dispatched cannot be recalled. Related copies follow their own schedules or require a separately scoped purge.
- A scoped live-data purge does not automatically delete provider logs, BigQuery usage records, or Firebase authentication accounts.
- Backups, PITR history, and preexisting soft-deleted copies follow their separate recovery and expiry rules.
- Copies already delivered to a customer, callback destination, or other customer-directed recipient require action by that recipient.
- Aggregate usage and account records require a separate retention or deletion decision; document deletion alone does not remove them.
- Restricted purge and deletion-control metadata can remain to record and manage deletion activity, without document bodies by design.
Where a request extends beyond the available purge scope, we identify the additional systems or recipients requiring separate action and coordinate as appropriate under the applicable agreement. Deletion from those systems is not automatic. Applicable legal requirements and contractual obligations govern request handling and completion timelines.
6. Retention decisions and provider boundaries
Some records may need to be retained to comply with applicable legal obligations, resolve disputes, or address justified security and abuse-prevention needs. Retention must be limited to the specific records, purpose, and necessary duration, with access restricted appropriately. Customer Personal Data covered by the DPA remains subject to its retention and deletion requirements.
Provider retention depends on the selected service, account settings, and contracts. BYOK does not eliminate Nugget's processing or the provider's retention, and Document AI processing is separate from LLM credentials. A Nugget live-data purge does not automatically delete provider-held copies. Provider obligations and international-transfer arrangements are governed by the DPA and applicable provider contracts; see the provider register.