Per-table child-object aggregates (reads Elasticsearch - listing discipline; never use for money decisions). Metrics: people_count, companies_count, people_with_linkedin_count are "now" totals; people_added_count, companies_added_count are bounded by the required period (half-open [from,to) over the children's created_at - attach time proxied by creation). filter scopes which TABLES aggregate (default live rows).

Two grouping modes, mutually exclusive (both set → 422): group_by:"sid" → groups[] with one {key, counts, metrics} entry per playtable (dashboard per-table numbers; cap 500 groups - narrow filter.sid.in); group_by_field_sid:"pt_fd_…" → field_value_groups[]: distribution of the field's object_type rows across the options of one scalar_select field (stage funnel; null key = empty/unset cell). group_by_field_sid requires filter.sid.eq = the field's playtable and a scalar_select field (422 otherwise).

When NOT: per-status people breakdown inside one table → search_people counts with filter.playtable_sid; listing/editing rows → people / companies tools.

Contract:

• MCP tool get_playtables_metrics, registry package mcp.playtables/playtables, mount playtables.tables.
• Operation metrics, response envelope metrics.
• Flags: read only.