D1: Classification Scheme

This document covers the schema design, priority logic, confidence framework, keyword dictionary, and example coding scenarios for the NEPA trigger classification pipeline.

For class definitions and NLI hypothesis statements, see Trigger Classification.


Suggested data structure

A project-level schema could look like this:

Variable Type Description
nepa_trigger_primary string Main federal nexus category
nepa_trigger_secondary list/string Additional trigger categories present
nepa_trigger_multi list/string All categories detected
nepa_trigger_evidence_text string Short snippet supporting classification
nepa_trigger_evidence_source string Where evidence came from
nepa_trigger_confidence integer or string Confidence in classification
nepa_trigger_manual_review boolean Whether human review is recommended
nepa_trigger_notes string Analyst notes

Primary vs. secondary trigger logic

Many projects have multiple forms of federal involvement. For example: - A transmission line may cross federal land and require a permit - A transit project may receive federal funding and require a USACE permit - A renewable project may be on BLM land and involve a federal right-of-way grant

Rule: assign the primary trigger as the federal nexus most directly responsible for the agency’s NEPA decision.

Priority order for primary trigger assignment (as implemented)

Use this only when the text is ambiguous. This is the order implemented in TRIGGER_HIERARCHY in 01_extract_nepa_trigger.py:

  1. federal_program
  2. federal_direct_action
  3. pma (Power Marketing Administration (PMA) + Tennessee Valley Authority (TVA))
  4. federal_property_transaction
  5. federal_land
  6. federal_permit
  7. federal_funding
  8. unknown

Why this order: A programmatic umbrella outranks everything because it changes the level of the NEPA review itself (see the example below). Below that, priority goes to the federal role when the agency is the principal actor or decision-maker, rather than simply one source of support or one regulatory checkpoint. PMA/TVA projects are elevated above land and permit because the agency identity is the clearest nexus signal for those entities.

Note (updated 2026-07): the ordering is no longer empirically inert for federal_program — 17 projects in the current published output carry federal_program together with another class in nepa_trigger_multi (7 + federal_funding, 4 + federal_land, 3 + federal_direct_action, 3 + federal_permit), and in all 17 cases the hierarchy determines the primary (federal_program wins).

Examples: - If BLM is approving a right-of-way across federal land, federal_land is often more informative than federal_permit - If DOE is building or directly implementing the action, federal_direct_action should outrank federal_funding - If the document is a programmatic plan, federal_program should outrank narrower project features - If BPA, WAPA, SEPA, SWPA, or TVA is the lead agency, assign pma as primary even when the project also crosses federal land or requires a permit


Keyword dictionary

Keywords are a first-pass filter only — see the “Important caution” note below.

federal_funding

Indicative terms: federal funding, federally funded, federal financial assistance, grant, cooperative agreement, loan guarantee, cost share, reimbursable agreement, funded through

Agencies often associated: DOE, DOT, HUD, EPA, USDA, FEMA

federal_land

Indicative terms: federal land, public lands, right-of-way, ROW grant, special use permit, easement, crossing federal land, National Forest System lands, BLM land, Bureau of Land Management land, Forest Service land, military reservation

Agencies often associated: BLM, USFS, NPS, Bureau of Reclamation, DoD, USACE (when property control is central)

federal_permit

Indicative terms: permit, license, authorization, approval, certification, permit application, jurisdictional waters, Section 404, incidental take permit, license amendment

Agencies often associated: USACE, FERC, FAA, FCC, NOAA/NMFS, USFWS

federal_direct_action

Indicative terms: agency proposes to, the Bureau proposes, the Department proposes, federal action consists of, construct, install, upgrade, operate, implement, federal facility, base operations

Agencies often associated: DoD, VA, USFS, BLM, Bureau of Reclamation, DOE, USACE

federal_program

Indicative terms: programmatic EIS, PEIS, site-wide EIS (SWEIS), Tier 1 review, policy, rulemaking, integrated resource plan, program-wide, nationwide, regional plan

Note: land-management programmatic reviews (e.g., vegetation management PEAs, leasing program PEIS on federal lands, BLM wind/solar PEIS, Western Solar Plan, Section 368 corridor PEIS) are classified as federal_land, not federal_program.

pma — Power Marketing Administration (PMA) + Tennessee Valley Authority (TVA)

Indicative terms: Bonneville Power Administration, BPA, Western Area Power Administration, WAPA, Southeastern Power Administration, SEPA, Southwestern Power Administration, SWPA, Power Marketing Administration, Tennessee Valley Authority, TVA

Agencies: BPA, WAPA, SEPA, SWPA, TVA, and any generic PMA. Assign pma as the primary trigger whenever one of these entities is the lead or sponsoring agency, even when the project also involves federal land (e.g., transmission line ROW grants) or permits. Add federal_land or federal_permit as secondary triggers when applicable.

federal_property_transaction

Indicative terms: land exchange, conveyance, disposal, transfer, sale of federal land, acquisition, parcel transfer

Important caution

Do not classify solely on keyword presence. For example: - “permit” may appear in a background section even if the true trigger is federal land - “funding” may be mentioned historically without being the current federal nexus - “right-of-way” can be central or merely descriptive

Recommended rules engine: 1. detect candidate categories 2. extract surrounding evidence text 3. rank likely categories 4. assign confidence 5. send low-confidence cases to review


Confidence framework

High confidence

Assign high when: - the text explicitly states the federal action or approval that triggers review - multiple sources agree - the language is direct and project-specific

Example: “The Bureau of Land Management must approve a right-of-way grant across federal lands” → high confidence federal_land

Medium confidence

Assign medium when: - the project strongly implies a trigger - one source gives clear but indirect evidence - multiple candidate triggers exist but one appears most likely

Example: transmission project with BLM lead agency and repeated ROW language, but no explicit sentence stating the nexus

Low confidence

Assign low when: - the project is clearly under NEPA review but the trigger is unclear - only weak keyword matches are found - several categories are plausible with no dominant one

Flag these for review.


Example coding scenarios

Example 1: Transmission line across BLM land

Evidence: Project title mentions new 230-kV transmission line; description says the project crosses BLM-administered land and needs a right-of-way grant.

  • nepa_trigger_primary = federal_land
  • nepa_trigger_secondary = federal_permit only if a separate federal permit is clearly required
  • confidence = high

Example 2: Wetland fill permit for private development

Evidence: Private developer; USACE permit under Section 404 required.

  • nepa_trigger_primary = federal_permit
  • confidence = high

Example 3: Federally funded transit improvement

Evidence: DOT grant funding is central to the project; no federal land use mentioned; NEPA document tied to funding approval.

  • nepa_trigger_primary = federal_funding
  • confidence = high

Example 4: Forest Service vegetation management project

Evidence: Forest Service proposes thinning and fuel reduction on National Forest lands.

  • nepa_trigger_primary = federal_land
  • nepa_trigger_secondary = federal_direct_action optional if you want to separately record that the agency is the direct actor
  • confidence = high

Example 5: Programmatic leasing framework

Evidence: Document is a Programmatic EIS for a regional leasing strategy.

  • nepa_trigger_primary = federal_program
  • confidence = high

Example 6: BPA transmission line with ROW on federal land

Evidence: Bonneville Power Administration proposes to rebuild a 230-kV transmission line; project requires a right-of-way grant across BLM-administered land.

  • nepa_trigger_primary = pma
  • nepa_trigger_secondary = federal_land (ROW grant on BLM land recorded as secondary nexus)
  • confidence = high

Note: pma takes priority over federal_land because BPA is the lead agency and primary actor. The land nexus is real but secondary.

Example 7: Ambiguous energy project

Evidence: DOE appears in metadata; document mentions funding and approvals but not clearly; no explicit statement of the federal nexus.

  • nepa_trigger_primary = unknown or best inferred category
  • confidence = low
  • manual_review = TRUE

Implementation options

Minimal viable

Start with just these variables: - nepa_trigger_primary - nepa_trigger_confidence - nepa_trigger_evidence_text - nepa_trigger_manual_review

Expanded (for research use)

  • nepa_trigger_primary
  • nepa_trigger_secondary
  • nepa_trigger_multi
  • nepa_trigger_evidence_text
  • nepa_trigger_evidence_source
  • nepa_trigger_confidence
  • nepa_trigger_manual_review
  • nepa_trigger_notes
  • nepa_trigger_rule_id (which rule fired)
  • nepa_trigger_model_label (if using ML/LLM adjudication)