M3 · BOUN-11293·TOTony·fieldwork-dashboard, fieldwork-processing

F3.3 — Dashboard Metrics

Deliver the data layer for the WPP fieldwork dashboard (F1.6): an hourly per-market metrics snapshot produced inside the fieldwork-processing Step Functions workflow (EventBridge hourly start → discover → fail-closed QC gate → per-market metrics compute → S3). Each successful cycle writes a validated, aggregate-only MetricsSnapshot JSON to a dedicated snapshot bucket: There is no HTTP API in v1 (design change 2026-08-05): the previously planned /v1/dashboard/* endpoints, ?fresh=true recompute, OpenAPI client codegen, and per-market SQS fan-out are all gone. The dashboard reads the snapshot JSON directly from S3; the embedded computedAt is the UI's "last sync" indicator. There is no per-request recompute path. v1 is scoped to data we already capture: completes vs target, starts/in-progress/IR%, screen-out breakdown, per-QC-check counts, research-defender counts, quota-cell fill, LOI (average/median/per-section), daily completes, grade/region tallies, and per-cell drill-down data. Derived values (percentages, deltas, totals) are deliberately not stored — consumers compute them. Forecast/RAG/Fount go-live metrics are v2, blocked on WPP supplying field dates.

19 of 28 stories shipped across 31 merged fieldwork-processing PRs: the hourly pipeline (discover to fail-closed QC gate to per-market metrics compute to S3 snapshot) is wired end to end, the 17 QC methods and workbook builders are ported, and a per-client API-key metrics endpoint shipped. Two export-scope stories are blocked on open product decisions and five more remain todo, including streaming export artifacts to S3 instead of buffering them in memory.