Synthetic institution

The developer app uses existing records from tests/testthat/fixtures/designed_test_data.R. The adapter in dev/demo-data.R assembles those scenarios; it does not maintain a second invented population.

The demo is its own institution: its departments and colleges come from institution/demo/, never UNM’s files (ADR-002 Stage 5). Add a row there when a fixture introduces a new code.

The default export contains five copies of the complete histories. Cohort 1 keeps the original source identities. The production transformation pipeline creates the analytical tables and encrypts IDs consistently; provenance columns make each person and section traceable afterward.

Use it

bash scripts/dev.sh up generates the institution in isolated Docker volumes and serves it at http://localhost:3838/. Source changes invalidate the saved data automatically on the next up.

To export a portable bundle with the project’s prepared R environment:

Rscript dev/generate-demo.R output/synthetic-institution

The exporter refuses populated directories without a synthetic-data marker. The bundle includes source-report tables, all cedar_*.qs app tables, fixture-people.csv, fixture-sections.csv, and synthetic-institution.json. The JSON records actual row counts, terms, campuses, and scenario families.

Select the original records

students <- qs2::qs_read("output/synthetic-institution/cedar_students.qs")
original <- dplyr::filter(students,
  synthetic_cohort == 1L, fixture_source == "base")

# Original fixture ID, even though the app's student_id is encrypted:
dplyr::filter(original, fixture_student_id == "STU-HD-SP1-001")

fixture_source distinguishes base enrollment/population histories, Regstats, Gen Ed associations, campus movement, retention, graduates, and major-entry scenarios. The section table also includes Seatfinder and rotating topics. Use both cohort and scenario when checking original expectations: the complete institution intentionally combines several populations.

The base HIST 1110 Spring 2020 cohort retains 21 registrations and nine late drops: reconstructed census enrollment is 30. The BIOL 2305 Fall 2020 internal crosslist retains three CRNs and a deduplicated total of 71. NURS 2010 supplies waitlist demand; Regstats includes growing, declining, and flat histories.

What is adapted

Missing section metadata is completed from the fixture’s enrollment rows. People with no program history receive explicitly undeclared records; known students do not receive invented historical declarations. Section identifiers are unique across scenarios and copied cohorts. College codes, date fields, and fractional faculty appointments are converted to source-report conventions. Added sections are marked synthetic_completion. Enrollments, grades, program changes, and terms come from the fixtures.

The independently authored DESR snapshots and class-list headcounts can differ. The exporter does not invent students to force them to agree. Admissions records and saved projections are absent. Deliberately corrupt fixtures and standalone intermediate credit tables remain unit-test inputs.

Copied cohorts provide volume for exploring the app. They are not independent statistical observations, a performance benchmark, or institutional evidence.