数据留存
各类数据的保留时长，以及负责删除的任务。
这些政策以英文发布。英文文本为准，译文仅供参考。
最后更新: 2026-08-01
本服务运营公司的注册信息尚未公布。信息确认后将显示在此处及页脚。
Automatically deleted
A background job runs every six hours and deletes the following. These are the windows that are compiled into the running service, not targets.
| Data | Kept for | Measured from |
|---|---|---|
| Search execution logs | 180 days | Creation |
| Authentication tokens (password reset, email verification) | 30 days | Expiry, or consumption if earlier used |
| Directory admission decisions | 365 days | Creation |
| HS-code, top-buyer and buyer-preview caches | Until their own expiry timestamp; a job sweeps hourly | Per-row TTL |
| Application log files | About 31 daily files | Rolling file limit |
Kept until you delete it
The following is retained for as long as the account or listing exists, because deleting it would remove something you are still using. It is deleted on request, or when an account is closed.
- Your account record (name, email, password hash, plan, Google identifier).
- Your searches and their stored results.
- Payment records and NOWPayments callback payloads — these are also subject to statutory accounting retention, which overrides a deletion request for the amount and date of a transaction.
- Contact-form and listing-request submissions, including the IP address and user-agent captured with them.
- Documents uploaded to claim or verify a company.
- Website-scrape requests and their extracted suggestions.
- Company profiles in the directory, including profiles built from public sources.
To have any of it removed, write to privacy@shelftap.com, or, for a company listing, use the 删除或更正公司条目 form.
Deletion mechanics
Deletion is a hard delete from the operational database, not a soft flag. Because there is no published backup-restore commitment (see 安全概览), we also make no claim about deletion propagating to historical infrastructure snapshots held by the hosting provider.
Known limitation
The purge job deletes at most 2,000 rows per data type per run. If a category ever accumulates faster than roughly 8,000 rows per day, the oldest rows can persist past the stated window until the backlog clears. We publish this rather than describe the windows as absolute.