MSHA Mines - landscape for Facility Foundry v1 onboarding

Phase 1 desk landscape, written 2026-05-06. Reference document for the MSHA Mines fetcher built against the v1 multi-dimensional Classification schema. Cross-references: docs/adr/007-multi-dimensional-classification.md, docs/DATASOURCES.md, src/strata/fetchers/msha/, the official schema definition file at https://arlweb.msha.gov/OpenGovernmentData/DataSets/Mines_Definition_File.txt.

Live-data observations (verified 2026-05-06 against the 2026-05-01 bulk dump):

1. The bulk URL is https://arlweb.msha.gov/OpenGovernmentData/DataSets/Mines.zip (~7 MB; weekly Friday refresh per MSHA's Open-Government-Data programme). The companion Mines_Definition_File.txt (~10 KB) is the canonical schema dictionary. A sibling Mines.txt file served from the same directory is a stale legacy mirror (Last-Modified frozen at 2021-03-22) - always consume the ZIP.
2. The pipe-delimited file has malformed RFC-4180 quoting. The free-form DIRECTIONS_TO_MINE field carries unescaped internal double-quotes; quote-aware parsing silently merges ~58 % of rows. The fetcher reads with quote_char=None (literal pipes only) and strips the surviving outer quotes per cell in _values.clean_text. Verified row count after the fix: 91,746 (matches the published record count).
3. 51.6 % of rows have lat/lon; 48.4 % don't. Mostly historical Abandoned mines (78,009 of 91,746) lack coordinates because MSHA began capturing geolocation only on more recent inspections. Admission drops the no-coords subset; live first-fetch admitted 47,330 facilities. Phase 1.5 may add a county-centroid fallback when cross-source resolution against MRDS proves out.

This is the reference for the v1 MSHA fetcher: slim Facility + multi-row Classification + FacilityIdentifier + OperatorReference + LifecycleEvent + FacilityAttribute.

---

1. Source overview

Publisher. US Mine Safety and Health Administration (MSHA), an agency within the US Department of Labor. The Mines table is the authoritative US registry of coal + metal/non-metal mines under MSHA's jurisdiction - both currently active sites and historical (abandoned, sealed, idled) ones.

License. US federal-government work - public domain (17 U.S.C. § 105). No registration, attribution, or fees required; courtesy citation per MSHA's Open-Government-Data conventions.

Citation (canonical). U.S. Mine Safety and Health Administration (MSHA), Mines Dataset, Open Government Data programme. https://www.msha.gov/data-and-reports/mine-data-retrieval-system. Edition: 2026-05-01.

Public landing page. https://catalog.data.gov/dataset/msha-mines-dataset (data.gov catalog entry); the MDRS portal at https://www.msha.gov/data-and-reports/mine-data-retrieval-system houses the search UI + a 20-table flat-file collection.

Vintage. Weekly refresh, Fridays. The Mines table's Last-Modified HTTP header is the de-facto vintage stamp; the fetcher records it as edition_tag. As of 2026-05-06 the latest is 2026-05-01.

Coverage. ~92K rows (91,746 verified 2026-05-06): 13,737 currently operating-ish (Active + Intermittent + Temporarily Idled + NonProducing + New Mine) plus 78,009 historical (Abandoned + Abandoned and Sealed). 56,024 Metal/Non-Metal vs 35,722 Coal.

---

2. File-level structure

The ZIP. Mines.zip (~7 MB). Contains one pipe-delimited text file Mines.txt (~37 MB uncompressed). Header row carries 60 columns in the order described in §3.

Encoding. ASCII text with CRLF line endings. Some rows carry non-ASCII bytes (apparent CP-1252 → UTF-8 mismatches) inside the DIRECTIONS_TO_MINE field; the fetcher reads with encoding='utf8-lossy' to substitute U+FFFD without losing rows.

Quoting. Text cells are wrapped in double quotes ("…"); numerics

• empties are bare. However, DIRECTIONS_TO_MINE carries unescaped internal quotes. Standard RFC-4180 readers (polars quote_char='"', pandas default) silently merge ~58 % of rows. The fetcher disables quote interpretation entirely (quote_char=None) and strips the surviving outer quotes per cell in _values.clean_text.

Sibling tables. MSHA's MDRS programme publishes ~20 flat files (Inspections, Accidents, Violations, Quarterly Employ/Prod, Annual Employ/Prod, etc.) keyed on the same MINE_ID. None are consumed here - the Mines table alone covers FF's identity / classification / operator needs. Phase 2 may pull Annual Employ/Prod for production time series enrichment (analogous to MRDS' Production_detail).

---

3. Column-by-column schema dump

The bulk Mines file publishes 60 columns. The fetcher extracts a subset (~30 columns) - the rest are operational metadata (inspection-office assignment, hazard-classification dates, free-form directions) that don't fit the FF schema cleanly. The "v1 destination" column points each consumed MSHA field to where it lands.

3.1 Identity + classification anchors

3.2 Geographic location

country_iso2 is hard-coded to "US" - MSHA's jurisdiction is US-only by statute.

3.3 Industry + commodity

3.4 Operator + controller

entity_id follows the lineage convention <scheme>:<value>: msha_legal_entity_id:<id> when MSHA's internal entity ID is populated; msha_operator_name:<sha1[:16]> deterministic fallback otherwise.

3.5 Operational attributes (FacilityAttribute rows)

---

4. Live ingest baseline (2026-05-06)

rows scanned:            91,746
admitted:                47,330  (51.6 %)
dropped (no coords):     44,416  (48.4 %)
records persisted:      996,726
  facility:              47,330
  facility_identifier:   47,330
  classification:       238,957
  lifecycle_event:       47,279
  operator_reference:    94,316
  facility_attribute:   521,514

Status distribution among admitted facilities: 33,946 decommissioned (Abandoned / Abandoned and Sealed with coords), 12,157 operating, 1,176 suspended, 51 commissioning.

Top NAICS industry codes among admitted facilities: 22,712 212111 (Bituminous Coal Surface), 12,408 212321 (Sand & Gravel), 8,152 212312 (Stone Quarrying), 1,849 212390 (NonMetal), 1,208 212114 (Anthracite), 989 212230 (Metal Ores).

---

5. Cross-source overlap with USGS MRDS

A spot-check of Bingham Canyon coordinates (UT, ~40.55, ~-112.10): MSHA admits Bingham Canyon Mine + Kennecott Keystone Underground + Copperton Concentrator + Kennecott Barneys Canyon Mining + Jaw Crusher Cedar Rapids 3042 at the same complex. MRDS has a single Bingham Open Pit Mine (dep_id 10069296) at slightly offset coordinates (40.52077, -112.1458) - same physical site, different sub-facility granularity, different lat/lon precision.

This is exactly the cross-source identity-resolution scenario ADR 005 is designed for. Phase 1 ingest preserves both sources' identities (separate facility_ids minted by each fetcher); Phase 2 synthesis will reconcile via name + coord + state similarity. Until reconciliation lands, downstream consumers querying the resolved store see both records as separate facilities.

---

6. Known gaps / Phase 1.5 work

• County-centroid fallback for the 44K no-coords admissions. The ~78K Abandoned mines mostly lack lat/lon. A fallback that derives a county centroid from STATE + FIPS_CNTY_CD (with coordinate_confidence=PLACEHOLDER) would surface those as county-grain facilities - useful for "is there a historical mine in this county" queries even without point precision.
• Production time series. MSHA's separate Annual Employ/Prod table (keyed on MINE_ID) carries quarterly + annual production tonnage. Joining it would give MSHA-side annual_production_<commodity> attributes analogous to the MRDS-side ones.
• Inspection / violation history. Out of scope for the canonical store but a strong signal for facility-status changes; consider as a LifecycleEvent source (regulatory-action payload) in Phase 2.
• SIC long-tail commodities. The current SIC_TEXT_TO_COMMODITY table covers ~37 entries; long-tail SIC descriptions fall through with no commodity emission (industry NAICS still admits via PRIMARY_CANVASS_CD). Expand as needed when domain consumers surface uncovered values.