The problem
Two systems disagreed about what counted as a completed event. The snapshot pipeline used a stale five-status definition while the product recognized nine — so every downstream metric silently undercounted, for months.
The system
An audit that diffed pipeline definitions against the product’s source of truth, quantified the gap on production data, then fixed the canonical definition, enriched the schema, and backfilled history — with before/after counts published for every table.
How it's built
- Definition diff between snapshot jobs and application source of truth
- Read-only production verification before any write
- Idempotent backfill with per-table before/after reconciliation
Delivery
Investigated, fixed, and backfilled by one engineer in days — the kind of unglamorous correctness work dashboards depend on.
Results
- A 30-day activity metric corrected upward ~3.5× (hundreds of missed events)
- Every snapshot table reconciled against source-of-truth counts
- Definition now imported from one canonical constant — drift can’t recur silently
Client anonymized by industry. Detailed numbers and references available on a call.