Metric current_user lie (list full, KPIs zero)

The corpus prior

Agents treat metrics as ground truth for desk health. Dashboard culture and training data say KPI tiles summarize the lists below. When lists are populated but metrics are 0, agents still report “demo works” or “no data” inconsistently — or chase seed bugs that are not the real failure.

Wrong shape

my_pipeline:
  display: metrics
  aggregate:
    draft: count(SpendRequest where status = draft and requester = current_user)
# list region for same persona shows 5 drafts
# agent: “seed failed” or “metrics prove empty”

Static persona_homes residual can still be 0.

Right shape

1. Trust order: list/queue regions under session → stills / capture → metrics last.
2. On metric 0 + list > 0: assume aggregate/current_user materialization footgun until disproven (F10, #1629 G5) — not missing seeds.
3. Prefer list residual / live_desk / qa capture for demo bar.
4. Machine residual (#1632): product_quality / dazzle demo quality report metric_list risk= when metrics aggregates use current_user and sibling lists have seed hits (trust order still lists → stills → metrics). metric_list residual= only when seed-level metric filters score 0 while lists have hits (true disagreement). Risk does not thrash residual_total.

Why this matters here

1629 Spend Desk: employee pipeline metrics all 0 while draft lists had rows.

Cognition that “metrics are truth” produces false beliefs even when seed and lists are correct.