ferpa-config.yaml

# FERPA audit configuration (repository root)
# Referenced by .claude/skills/ferpa-audit/scripts/* — keep stable keys for hash comparisons across runs.

version: 1

project:
  dashboard_relative_path: codebenders-dashboard
  analyze_route_glob: codebenders-dashboard/app/api/analyze/route.ts
  execute_sql_route_glob: codebenders-dashboard/app/api/execute-sql/route.ts
  ai_transparency_file: codebenders-dashboard/content/ai-transparency.ts
  query_executor_file: codebenders-dashboard/lib/query-executor.ts
  config_file: codebenders-dashboard/lib/config.ts

# Columns that must not appear in SELECT lists, API JSON shapes, exports, or logs
# without an explicit same-line annotation: // FERPA-OK: <reason>
select_exclusions:
  - Student_GUID
  - student_guid

# Demographic fields — raw values in small cells heighten re-identification risk (Layer B)
sensitive_demographics:
  - Race
  - Gender

subpopulation_minimum_n: 10

# Postgres (Layer B) — override with DB_* env vars; see operations/db_config.py
database:
  schema: public
  predictions_table: student_level_with_predictions
  small_n_dimensions:
    - '"Race"'
    - '"Gender"'
    - '"Cohort"'

# Third-party / non-institutional hosts (Layer A fetch classification)
external_data_hosts:
  - hostname: schools.syntex-ai.com
    note: "Analysis-ready student-row API; must be blocked when FORCE_DIRECT_DB=true"

allowlisted_external_hosts:
  - api.openai.com
  - api.anthropic.com

rbac:
  header_name: x-user-role
  # API routes (relative to dashboard) that return student-level or row-level education data
  # and must enforce the role header per institutional policy.
  # execute-sql and analyze are covered by dedicated Layer A checks (FERPA guard + vendor prompt).
  student_data_routes:
    - app/api/students/route.ts
    - app/api/students/[guid]/route.ts
    - app/api/dashboard/kpis/route.ts
    - app/api/dashboard/readiness/route.ts
    - app/api/dashboard/retention-risk/route.ts
    - app/api/dashboard/risk-alerts/route.ts
    - app/api/query-history/route.ts

taxonomy:
  pii_column_substrings:
    - GUID
    - SSN
    - social
    - birth
    - email
  educational_record_tables:
    - student_level_with_predictions
    - course_enrollments
    - student_predictions
    - course_predictions

audit_log:
  # When set, Layer B will check for this table; if missing, emit informational Note only.
  table_name: null

llm:
  # Files that must stay aligned with ferpaExcluded in analyze_route (regex extract in Layer A)
  sdk_markers:
    - "@ai-sdk/openai"
    - "createOpenAI"
    - "streamObject"
    - "generateText"