The freshness monitor watched defunct nspd_scrape_runs (3 WAF-banned manual
runs 2026-04-30, triggered_by=manual, never scheduled) → false "no successful
runs" alert. Investigation (surfaced by the monitor's first prod run) showed
NSPD is healthy: the live harvest (tasks.nspd_sync.harvest_stale_quarters +
the #1891 /analyze resolver write-back) writes nspd_quarter_dumps directly —
370 rows, 56 in the last 7 days, last fetched 2026-06-22 — but never writes
nspd_scrape_runs.
Add a data-table freshness mode (FreshnessSource.timestamp_col): when set, the
source uses MAX(timestamp_col) instead of run-ledger status semantics. Repoint
nspd → nspd_quarter_dumps.fetched_at_utc (fresh<14d, stale<30d). Run-ledger
sources (kn/objective/nspd_geo/cadastre) untouched. No schema/codegen change.
Refs #73
