Public Q&A library shipped without leaking internal identifiers
Rebuilt an open knowledge library so anonymous readers see only vetted answers, while every internal user identifier stays server-side. Column- and row-scoped policies enforce it at the database.
Challenge
The prior library exposed all rows to anonymous readers, including internal author UUIDs. Any scraper could correlate authorship across public content — an unacceptable privacy posture for a Trust-Center-backed platform.
Solution
Replaced the open policy with a column allow-list (no author identifier reaches the wire) plus a row filter that admits only vetted content (quality score ≥ 0.7). Documented the rule in the security memory so future scans and future agents inherit the constraint.
Outcome
Zero internal identifiers reachable via anonymous read. Draft and low-quality content invisible to the public. Trust-Center privacy claims independently verifiable via database policy inspection.