discovery_signals
Agency activity spikes
Per-agency document-output spike signal that powers the feed's discovery panel — agencies whose document count in the trailing 30 days is at least 2× their prior-year monthly mean, requiring ≥24 documents in the prior year to suppress tiny-denominator noise. Recomputed relative to the run date, so rows turn over on every rebuild; it is a current-state signal, not a time series. Built by build_discovery_signals from documents, less postings Regulations.gov removed (documents.publisher_status).
Coverage. Not a range. A dated snapshot recomputed from documents at each build. Its date is the calculation instant the file's metadata records (spicy_regs.discovery_signals.as_of, beside the date policy and the documents digest it read; parquet_kv_metadata on the generation's file returns them), not measured_on, which dates the audit of this note. Future dates are excluded from recent activity. This is not a historical series or a complete measure of agency activity. Receipt: fork-execution-2026-09-21/discovery-utc/MANUAL-AUDIT.md. (measured 2026-09-21)
Data quality. Activity windows use UTC instants and preserve explicit source offsets. Dates and timestamps without offsets use UTC; source values remain unchanged. Corpus backfills and source date anomalies can affect these document-count signals.
- Parquet file:
discovery_signals.parquet - MCP
query_sqlsupport: Configured; requires an available artifact. - Publication status: Not established by this schema page or its measurement date.
- Row count: Not stated here; the MCP
describe_tablereply gives the live count underpublication.
| Column | Type | Description |
|---|---|---|
agency_code |
VARCHAR |
Agency short code. |
recent_30d |
BIGINT |
Documents the agency posted in the trailing 30 days. |
baseline |
DOUBLE |
Prior-year monthly mean document count (months 2–13 back). |
ratio |
DOUBLE |
recent_30d divided by baseline — the spike multiple. |