Get Performance Combined Read

Per-day series behind the two headline number cells (the sparklines).

active_users_series is converted_contacts per day bucket, avg_session_length_series the average measure — both zero-filled over one shared day axis (oldest to newest), so the lists share one length. series_provisional_last flags the trailing still-accumulating bucket. When not_yet_enabled every list is empty (no randomized base to chart).

One (metric, audience, region) arm extrapolation in the combined read.

The per-scope projection every PerformancePage projection surface consumes: the stats cell reads primary_projection; each tracked-target chip reads the projections entry matching its stored (metric_event_key, audience, region_id) triple. Per arm the population *_n and the distinct in-window converters *_converters yield a rate (None at n == 0, distinct from a measured 0.0); lift = (treated_rate - control_rate) / control_rate (None unless both rates and control_rate > 0); projected_additional_converters = (treated_rate - control_rate) * control_n (None unless both rates; unclamped) — the combined read folds every chip's projection here instead of the page firing the projection route once per chip. not_yet_enabled lives on headline_metrics, not here: a not-yet-enabled org's entries are the zero/None cells the stats cell (unmounted) never consumes.