spec: 22 more operations (community, birdwatch, lists, explore, viewer) and tonight's drift

[fa0311]

Adds the 22 placeholder operations that had a recorded response but no spec path (64 → 86 operations). Every response schema is typed from the recorded response (required = present in the sample); the strict live test now calls all of them with two cursor pages.

New operations

• community (new tag → CommunityApi): CommunitiesFetchOneQuery (community page), CommunityQuery (settings / moderation pages), CommunitiesSearchQuery (cursor in slice_info.next_cursor), TopicCarouselQuery, GlobalCommunitiesPostSearchTimeline, GlobalCommunitiesLatestPostSearchTimeline.
• birdwatch (new tag → BirdwatchApi): BirdwatchFetchAliasSelfSelectStatus, BirdwatchFetchAuthenticatedUserProfile, BirdwatchFetchBirdwatchProfile, BirdwatchFetchGlobalTimeline (segmented timelines), BirdwatchFetchPublicData (public data file urls).
• tweet: BookmarkSearchTimeline, ExploreSidebar. user-list: ListSearchTimeline, CombinedLists (same UserTweetsResponse shape as ListOwnerships), ListsManagementPageTimeline. users: UsersVerifiedAvatars. default (paths/viewer.yaml): AuthenticatePeriscope, CreatorStudioTabBarItemQuery, DataSaverMode, GrokHome, PinnedTimelines.
• Not added yet: getAltTextPromptPreference, isEligibleForVoButtonUpsellQuery, usePremiumPaywallOnLoadMutation (their placeholder keys start lowercase, which the strict harness's method-name → key mapping does not handle), and the catalog-only operations that have no placeholder entry.

Relay community pages

The three community queries are Relay fragments: node ids on the result wrappers and on rules, __is… interface markers on the action results, and partial users (facepile users are {__typename, avatar, id}, the settings creator is {__typename, core.screen_name, id, is_blue_verified, verification}). They get their own types (CommunityPageCommunity, CommunitySettingsCommunity, CommunitySearchCommunity, CommunityFacepileUser, CommunityRelayRule, CommunityPageActions, …) instead of widening Community / User; only the objects that are the same at every site are shared (CommunityBannerMedia, CommunityViewerRelationship, UserAvatar, AffiliatesHighlightedLabel, Timeline / TimelineResult). UsersVerifiedAvatars likewise returns a partial VerifiedAvatarUser. TypeName gains CommunityHashtagSlice and CommunityListMembersAction.

Placeholder / tooling

• tools/update_placeholder.py now fills in variables for an entry that has none (X answers 422 without them); that completed CommunitiesFetchOneQuery, UsersVerifiedAvatars and isEligibleForVoButtonUpsellQuery from the stamped catalog lines. HomeTimeline is POST per the latest stamped catalog line.
• The entries that carried the identifiers of the account that was logged in during the earlier recordings (ListOwnerships, UsersVerifiedAvatars, GenericTimelineById, ProfileSpotlightsQuery, Likes, isEligibleForVoButtonUpsellQuery) were re-recorded today on the sandbox account and their sample values set to the usual ones (screen_name elonmusk, userId 44196397, tweetId 1349129669258448897; the request shapes are the recorded ones). The same id is removed from the UserTweets id list of the strict test. catalog: re-record the viewer-bound requests on the sandbox account twitter_api_safe_relay#38 replaces the matching catalog rows.
• tools/build_config.py registers the two new tags; test/python/test_serialize.py iterates CommunityApi and BirdwatchApi too, and treats X's partial errors beside complete data (match rate 1.0) as a warning instead of a failure, which is what CombinedLists, ListSearchTimeline, ListsManagementPageTimeline and ListByRestId always return (the ListByRestId skip is gone). tools/json_to_schema.py (local, not committed) drafted the schemas from the recordings; everything was reviewed and renamed by hand.

Drift found by the strict runs (X side, 2026-10-16)

• ClientEventInfoDetails.timelinesDetails is optional: notification items carry notificationDetails {impressionId, metadata} instead.
• RichMessage.entities on notifications: TimelineRichTextUser refs with user_results and TimelineRichTextList refs with id and a relative url.
• ModuleFooter.displayType can be an empty string, so it is a plain string now.
• AdMetadataContainer.unifiedCardOverride (a JSON string) on a promoted tweet.
• Promoted trends: TimelineTrend.description / promoted_metadata (TrendPromotedMetadata: advertiser, impression ids, promotedTrend), TrendMetadata.meta_description; FeedbackAction.feedbackType gained Dismiss, so it is a plain string now.
• New item type TimelineEventSummary (promoted event card on ExplorePage: event id, hero media with a tweet, title, its own promoted metadata whose ad container lacks renderLegacyWebsiteCard), added to ItemContentUnion, ContentItemType and TypeName.

Verification

• 422 stored responses deserialize at match rate 1.0 (the 22 new operations and every response of tonight's strict runs among them).
• Strict live test with the nightly settings on the sandbox account: 125 calls, every match rate exactly 1.0, exit 0 (8 responses carried X's partial errors beside complete data and are logged as warnings).

🤖 Generated with Claude Code