Approve Experiment Review

Unique ID of the review.

Lifecycle status of the review.

Kind of change the review proposes.

The user who created the review.

Requested reviewers (users and groups).

Human-readable description of the proposed change.

ISO-8601 timestamp of when the review was created.

The proposed change this review would apply, as { current, proposed } over exactly the fields committing the review would write. current is the live value read at request time; where commit treats an unset argument as "leave unchanged", proposed repeats the live value rather than a default. Populated on single-review responses only — list responses omit it so a page of reviews does not pay an extra read per row. When the review does propose something but it could not be read back, content is PRESENT with unavailable_reason set and both value objects empty, never null — so "this type proposes nothing" stays distinguishable from "we cannot tell you what this would do". This surface's reasons are review_data_unavailable (the stored change data is missing or does not match the type the review declares; expected on legacy rows) and content_unreadable (the data is present but the before/after could not be assembled; a logged server-side fault). null when the review type is a bare action with no recoverable proposed value: stop, pause, restart, reallocate, delete, disable_reviews_locally (always), start (unless the review carries a backdated start date that commit will actually write — analysis-only experiments on warehouse-native projects), scheduled_start_edit when new_time is ignore (commit is a no-op), and update_settings on an autotune (the stored settings have no experiment-contract shape). Every other type carries content, including archive, unarchive, update_owners, update_team, change_enabled_groups and update_allowed_reviewers, for which the Console currently shows no before/after. Keys by type: archive/unarchive -> archived, allocation (archiving a layer-bound experiment zeroes its allocation); abandon -> decision_note_title, decision_reason; start -> start_time; scheduled_start/scheduled_start_edit -> start_time; rollout -> group_id, rollout_percentage, rollout_phases, has_upcoming_automated_rollout; schedule_rollout -> group_id, rollout_phases; make_decision -> winning_group_id, decision_note_title, decision_reason, and on experiments overrides, id_overrides, inline_targeting_rules_json, targeting_gate_id; update_overrides -> overrides, id_overrides, allow_overrides_in_pulse; update_target_applications -> target_app_ids; update_default_impact_multiplier -> default_impact_multiplier; update_owners -> owners; update_team -> team_id, allowed_reviewer_user_ids, allowed_reviewer_group_ids (moving to a team whose approval mode is admin_only/team_only widens who may approve future reviews); update_allowed_reviewers -> allowed_reviewer_user_ids, allowed_reviewer_group_ids; change_enabled_groups -> groups (the full list, with the delta applied); update_settings -> a single settings object holding the full resolved settings set, including overrides, idOverrides, allowOverridesInPulse and customScorecardSections.