atlas-ops-analytics

name

atlas-ops-analytics

description

Solve Atlas Commerce Operations analytical and controlled-correction tasks that provide a prompt, request payload, JSON answer contract, schema/data dictionary access, and SQL-style workplace data. Use for fulfillment, refund, carrier-scan, warehouse productivity, support health, or similar operational scorecards requiring exact JSON output, event-derived state at a cutoff, imported-row deduplication, FX conversion, risk classification, ranked exceptions, or approved canonical data corrections.

Atlas Ops Analytics

Core Workflow

1. Read the prompt, request payload, and answer template before querying data. Treat the payload as the business contract and the template as the output contract.
2. Inspect the schema and data dictionary supplied by the task. Prefer canonical fields for analytics; use raw fields only for source-reconciliation checks.
3. Build the cohort first, then calculate metrics from that cohort. Apply production filters from account flags when the request says production accounts, production orders, production shipments, or production customers.
4. Use UTC timestamp boundaries exactly as stated. Respect inclusive, exclusive, and "strictly before" language.
5. Deduplicate imported event or attempt rows before aggregation when source identity fields can repeat. Use (source_system, external_event_id) with latest ingested_at, then stable row id as a tie-breaker, unless the business entity has a stronger logical id such as refund_id.
6. Derive state at a cutoff from effective event history when event tables and snapshot columns disagree or when the request says "effective", "at cutoff", "active at cutoff", or "completed by cutoff".
7. Calculate rankings and classifications from unrounded values. Round only final reported numbers to the requested precision.
8. Validate the final object against the answer template: exact fields only, required array lengths, stable sort order, numeric precision, and no narrative.

Cohorts And State