Overrides

Every override is work our product should have done — fewer overrides, more joy

snapshot

Range

Override burden per day

Overrides on packages submitted that day, split by inclusions (deposits pulled into income) vs exclusions · line = overrides per processed package (right axis, the joy metric — down is good)

Zero-touch rate

The north star, inverted: share of processed packages with no human intervention — no override and no under-review stop · dashed = trend · drive this to 100% · data since Apr 13 (state-history floor)

What touches packages

Every processed package classified by intervention: none (zero-touch), overrides only, under-review only, or both · stacked per submission day · since Apr 13

Override burden by customer

Sorted by overrides in the selected range · Δ compares to the prior equal-length window · high ov/pkg = a customer doing our work for us

What the overrides are about

Described overrides keyword-bucketed into themes, by override date · only ~8% of overrides carry a description — themes are a sample of the whole

Automation opportunities — by theme

Trailing 180 days, grouped by theme with the customers who need each rule · direction consistency is the decision: consistent themes are global-rule candidates, mixed themes carry contradictory signals and need tenant scoping or extra conditions · ⚠ flags a customer whose direction contradicts the others

Methodology. Overrides from overrides_override: type 1 = inclusion (deposit pulled into qualifying income), type 2 = exclusion. Daily and per-customer metrics attribute a package's overrides to its submission day (cohort view) and count processed packages only; themes and opportunities attribute by override date. Zero-touch = a processed package with no overrides AND no status-3 (under review) row in packages_historicalloanpackage; its window starts 2026-04-13 (earlier state history is backfill artifacts) and support tickets are not part of the definition (no reliable package-level join). Only ~8% of overrides carry a free-text description; themes bucket those by keyword (first match wins) and raw description text is never shown — it can contain borrower-adjacent details. Internal and test tenants excluded (same list as the Revenue Report); weekends (Sat/Sun, Eastern) excluded from every chart and metric; chart lines smoothed with a centered moving average (3-day at 30d, 5-day at 90/180d; none at 7d) while KPIs and tables stay raw. Opportunities are grouped by theme over the trailing 180 days and ranked by described-override volume; per-package concentration is the tell — hundreds of overrides on a handful of packages means one repetitive judgment an auto-rule can absorb. Direction decides the rule shape: a theme is consistent when ≥85% of its overrides go one direction (global-rule candidate) and a mixed signal otherwise (tenant-scoped rules or extra conditions). Customers with ≥25 described overrides in a theme are listed; ⚠ marks a customer whose majority direction contradicts the theme majority, ⇆ marks a customer internally split ≥25/75 (conditional judgment). Rolling 180 days · laminr-prod.