Expand description
ai.hyprstream.event.group / ai.hyprstream.event.groupItem — lexicon
records for EventService managed-list groups (epic #600, EV2 #602).
Built on the generic crate::list_record module: a parallel,
pattern-matching lexicon, deliberately modeled on the signed-off
ai.hyprstream.placement.group/groupItem pattern from #524 (atproto
list, owner-curated, with a separate listitem per member; bidirectional
consent = owner lists the member and the member’s own record consents) —
but NOT a literal reuse of ai.hyprstream.placement.group, since that
lexicon is sign-off-frozen (unknown fields rejected at decode) and event
groups need an extra keysetId field placement groups don’t have.
Resolved design decision (human review, #602): keep these as separate
NSIDs/collections — this preserves cheap collection-based firehose/AppView
filtering and keeps placement-membership and event-key-membership as
genuinely distinct axes — while sharing the list/listitem/consent
mechanics via crate::list_record so neither lexicon hand-duplicates
validation, encoding, or the consent check.
§Key material is NOT in the record
keysetId is a stable opaque reference (“this group’s key material is
generation X”), never the key bytes. The actual K_group symmetric key is
generated fresh per rotation (exactly as SecureEventPublisher::register_prefix
does today) and delivered per-member via DH-wrap-on-join
(derive_wrap_key/wrap_group_key) — never committed to the public,
firehose-synced atproto repo. This matches #524’s “durable facts = atproto
records, volatile/secret facts = ephemeral RPC” split, and keeps the
relay/firehose blind to K_group (epic #600’s threat model).
§Member-side consent lexicon — still open
Membership is bidirectional: EventGroupItemRecord is the owner’s
claim that a member belongs to the group; it is NOT sufficient alone — the
member’s own identity record must independently consent (carry the group’s
at-uri in its own groups-style field). Which lexicon carries that
member-side consent for EventService members remains an open question:
candidates are reusing ai.hyprstream.placement.node.groups for members
that are also placement nodes, or a new minimal ai.hyprstream.event.member
record for members that are not. check_bidirectional_consent implements
the check generically over two already-fetched claims, deliberately not
committing to the member-side lexicon.
Re-exports§
pub use list_record::check_bidirectional_consent;
Structs§
- Event
Group Extra ai.hyprstream.event.group’s extra field beyond the common list shape.
Constants§
- GROUP_
COLLECTION_ NSID - NSID for the group (list) record.
- GROUP_
ITEM_ COLLECTION_ NSID - NSID for the groupItem (listitem) record.
Type Aliases§
- Event
Group Item Record ai.hyprstream.event.groupItem— the owner-side membership claim (listitem), lives in the owner’s repo. Identical shape to every other listitem lexicon in this crate — seeListItemRecord.- Event
Group Record ai.hyprstream.event.group— the list record, lives in the owner’s repo.