Track real eLumen + IR PeopleSoft schema capture for live-data enrichment

[johnnyrobot]

Goal

Complete the remaining real-data slice after m7 foundation.

m7 shipped the offline/deterministic plumbing:

• exact guarded DNF→CNF prerequisite conversion
• fixture-backed eLumen prerequisite mapping
• IR/PeopleSoft enrollment ingest
• detector activation when overlapping enrollment data is present
• engine.run() remains deterministic and offline

What remains is validating those seams against real source shapes.

Needed Inputs

1. eLumen prerequisite payload

Current assumed shape, not validated on real data:

{
  "source": "elumen ...",
  "campus": "LAMC",
  "courses": [
    {
      "course_id": "PHYS 102",
      "raw": "(MATH 245 and MATH 246) or PHYS 185",
      "dnf": [["MATH 245", "MATH 246"], ["PHYS 185"]]
    }
  ]
}

Required meaning:

• course_id: gated course in engine catalog form, e.g. PHYS 102
• raw: original human-readable prerequisite text for provenance/debugging
• dnf: outer list is OR branches; inner list is AND-required courses
• missing/empty dnf means no prerequisite

Need captured real response/schema before implementing a live HTTP client. Do not invent endpoint/auth.

2. IR PeopleSoft enrollment export

Expected .xlsx sheet: sections

Required columns:

• Term
• Class Nbr
• Cap Enrl
• Tot Enrl
• Wait Tot

Expected join: (int(Term), bare Class Nbr)

Notes:

• IR side Class Nbr should be a bare CRN, e.g. 20001.
• Live schedule side has decorated class_nbr, e.g. 17818 (LEC).
• Join strips the leading integer from live class_nbr.
• Real validation requires overlapping terms and CRNs with live schedule data.
• Instructor/student-level PII should not be included or committed.

Optional context columns may include:

Descr, Campus, Subject, Catalog, Section, Session, Class Type, Component, Class Status, Mode, IN_PERSON, Wait Cap, FILLD, .FILLPERCNT, ENRL, LMT, Acad Org, Dep, Discipline, IGETC, OER, FTE, Class Workload Hrs, LEVEL, CLASS, SEC, Class Start Date, Class End Date

Acceptance Criteria

• Real/captured eLumen schema is documented with sensitive data redacted.
• Real/captured IR export schema is documented with no student/instructor PII committed.
• eLumen adapter maps real payload into current course_id/raw/dnf internal shape.
• IR export with overlapping (Term, Class Nbr) activates enrollment-driven detectors.
• Zero-match enrollment joins remain explicitly reported as inert, not silent success.
• Prereq conversion flags guarded fallbacks when DNF→CNF expansion exceeds the clause limit.
• engine.run() remains offline/deterministic; all source fetching stays outside the engine.
• Tests cover real-shape fixture parsing, join match, zero-match behavior, privacy, and detector activation.

Non-goals

• No credentials in repo.
• No fabricated eLumen endpoint.
• No real student-level data.
• No networking inside engine.run().

Verification

python3 -m pytest -q

Expected result: all non-live tests pass, live tests remain deselected unless explicitly requested.

References (current state on main)

• Already shipped (fixture-only): sources/prereq_cnf.py, sources/elumen.py, sources/enrollment.py, sources/mapping.py, build_live_workbook.py
• Spec: docs/superpowers/specs/2026-05-30-m7-real-data-enrichment-design.md · Plan: docs/superpowers/plans/2026-05-30-m7-real-data-enrichment.md
• IR sample: files/lamc_sample_enrollment.xlsx (terms 2248/2252) · eLumen fixture: tests/fixtures/elumen_prereqs_LAMC.json
• Foundation PRs: feat(m7): add fixture-backed prereq and IR enrollment enrichment #14 (m7), fix(privacy): sanitize instructor PII in fixtures + correct N2 wording + real-key regression test #15 (instructor-PII sanitize + N2 wording)

🤖 Generated with Claude Code