Per-state Data Quality scorecard over the customer-facing lead_browse view.

Mechanical conformance checks across four value dimensions — format (state, phone, email, zip), completeness (registered agent, address), standardization (canonical vocabulary), and consistency (names in CRM-ready Title Case, not ALL-CAPS). Returns an overall 0–100 score, the per-dimension breakdown, and per-check pass rates with sample offenders.

residual_defects and checks_with_residual_defects ride beside the headline score: a small failure count (e.g. 159 defects over ~397K applicable rows) rounds away to a clean 100.0 at both the check and headline level, so these two fields are how a reader tells "zero defects" apart from "100.0 that is hiding a known residual" without digging into checks[].failed by hand.

The payload also carries the record-centric coverage dimension (the coverage block + dimensions.coverage): per pipeline stage, how many records that brain has never stamped — loaded-but-unprocessed inventory the value checks cannot see. Each coverage check ships browse_filters (a missing_stage filter) that opens the gap cohort on browse/export and scopes a surgical repair run on the pipeline trigger. Coverage does not move the headline score (see stage_coverage.py for the scoring rules).

The bad_emails block counts every known-bad email we hold (ZeroBounce invalid / spamtrap / abuse / do-not-mail, the email picker's own list): primary_bad_contacts (a bad email in the sellable slot — should be 0), set_aside_bad_emails (kept on purpose so we never pay to check them again), all_bad_contacts, and never_validated_contacts. Reported beside the score, never folded into it.

The bad_phones block is the phone-side mirror: primary_unknown_type_on_fire (an unresolved phone line type sold at full reachability_tier='On Fire' confidence — should be 0, the one competitor-mined defect this check exists to catch), plus primary_unknown_type_contacts and primary_not_confirmed_live_contacts for context (some residual there is normal — a real number with no vendor type/liveness signal).

This is the scoreboard the State GMs read each session and the State Quality Board renders, so data-quality work has a number to move the way classification does.