We currently have a de-normalized query cache. What does that mean? We have redundant data all over the place in the frontend (ReactQuery) cache.

Take a user profile as an example. It lives in many locations in our query cache:

• ACCOUNT — canonical [ACCOUNT, uid] query via queryAccount(). The primary source of truth for a single account's profile.

• LIST_ACCOUNTS — [LIST_ACCOUNTS] returns all known accounts, each with full profile metadata.

• DOCUMENT_COMMENTS — [DOCUMENT_COMMENTS, docId] embeds authors: Record<string, HMMetadataPayload> via loadAccounts() in createCommentsResolver().

• DOCUMENT_DISCUSSION — [DOCUMENT_DISCUSSION, docId] same pattern — createDiscussionsResolver() embeds author profiles.

• BLOCK_DISCUSSIONS — [BLOCK_DISCUSSIONS, docId, blockId] via createCommentsByReferenceResolver(), also embeds authors.

• COMMENTS — [COMMENTS, commentId] individual comment lookups that embed author profile.

• AUTHORED_COMMENTS — [AUTHORED_COMMENTS, uid] lists comments by an author, embeds profile data.

• ACTIVITY_FEED / FEED — each LoadedCommentEvent and LoadedCitationEvent embeds HMContactItem objects containing profile data (name, avatar) for all involved accounts.

• ENTITY — [ENTITY, id] for account-type entities, the resource itself contains the profile document with metadata.

• RESOLVED_ENTITY — [RESOLVED_ENTITY, id] same as ENTITY but after redirect resolution.

• CONTACTS_ACCOUNT — [CONTACTS_ACCOUNT, uid] contact records that include the contact's profile info.

• CONTACTS_SUBJECT — [CONTACTS_SUBJECT, uid] reverse contact lookups, also contain profile data.

• SEARCH — search results can include account matches with embedded profile metadata.

• CAPABILITIES — [CAPABILITIES, uid, ...path] capability entries reference delegate accounts whose profiles may be embedded in the response.

Why is that a problem? When some data changes- such as a profile blob comes in, we don't know what to invalidate. Because the profiles are scattered throughout, we basically need to invalidate the whole world when we see a new user profile come in. This results in extra-heavy data reloading.

Also, when data is duplicated everywhere inside the query cache, it leads to higher memory usage.