ANU Degree Planner

This is a full-stack replacement for the manual process of planning a degree against ANU’s Programs & Courses site (programsandcourses.anu.edu.au, “P&C”) — a system where a student opens a shareable plan, sees the Bachelor of Advanced Computing (Honours) requirement tree for the 2027 cohort starting Semester 1, drags courses onto an 8-semester timeline, and sees clearly, for every course, whether it’s available, blocked, and why — computed from a real recursive prerequisite check against P&C’s own course data, not a static checklist.

What good looks like here

Good means three things hold at once: the state shown for every course is never wrong (available/soft-blocked/hard-blocked, with a specific reason, not a guess); every mutation (a placement, a change to the completed semesters, a pin) is enforced server-side and persists exactly, so the client can’t drift from the source of truth; and the app never claims certainty about P&C data it can’t actually verify — an unparseable requisite clause shows as “verify on P&C” rather than being silently dropped or silently assumed true.

Most of that is enforced, not just aspired to: spec/invariants.test.ts runs axe and the CI deploy probes against every route; the domain layer (src/lib/domain/**) is unit-tested against real scraped course data, not fixtures invented to make the tests pass; and spec/*.test.ts drives the built server over HTTP, including the requisite parser, the allocation solver, and — for this phase — the runtime P&C fetch and search. What’s a judgement call rather than a checked rule: the exact wording of a suggestion (“Place X in S1 2028” vs. “Move X to…”), the visual design of the timeline, and which of two equally-defensible readings of an ambiguous P&C sentence the requisite parser takes.

Where this diverges from the original brief

The brief assumed a course could double-count toward more than one requirement group. This app replaces that with strict single allocation: each placed course counts toward at most one group, chosen by an optimal solver that maximises satisfied units, with pinning available when the student disagrees with the automatic choice. Double-counting would have let the same 6 units silently satisfy two different totals, overstating real progress toward the 192-unit program; single allocation is the one that can’t lie about how many units a student has actually earned toward the degree as a whole. Filter-based requirement groups (e.g. “12 units of 3000/4000-level COMP”) are supported in this v1, alongside explicit course lists.

Limitations

  • Offerings for 2029–30 are projected, not scraped: they copy the most recent year P&C actually published sessions for, and are visibly marked “projected” wherever shown. P&C simply hasn’t published that far ahead yet.
  • A course with no published offering at all is marked “unknown”, not hard-blocked — the app can’t tell the difference between “not running” and “not yet on P&C”, so it stays placeable with a “No published offering — verify on P&C” badge rather than refusing it outright.
  • TDP (Transdisciplinary Problem-Solving) tracking is untracked, not “0 of 12 units”: the scrape task spent a time-boxed hour searching for a machine-readable TD tag on P&C’s course search API and found none. No TD status is ever guessed, so the program check shows as “not tracked — verify on P&C” instead of a number that could be quietly wrong.
  • The Machine Learning specialisation was merged into ARIN from 2027, per P&C, so the specialisation options here are ARIN, HCCC, SYAR and THCS — an ML major from an earlier cohort year won’t appear.
  • The requisite parser is deliberately over-strict in at least one known case: COMP4350’s “either A or B and C” phrasing parses with the wrong precedence (unparenthesised OR is treated as binding tighter than AND, which matches most P&C phrasing but not this one). This is accepted rather than special-cased, because an over-strict parse only ever produces a false soft-block — never a false hard-block, and never a false “satisfied” — so the failure mode is “you may need to double-check this one on P&C”, not a wrongly-permitted enrolment.
  • P&C’s licence is unconfirmed. Scraped and live-fetched course data is republished here for a non-commercial student project, with attribution below; if that turns out not to be permitted, the data will need to come down.

Out of scope entirely for v1: authentication and accounts (a plan’s only protection is its unguessable URL); any program, cohort year or starting semester other than AACOM/2027/Semester 1; summer, winter and spring sessions; repeated or failed courses, part-time study, leave and credit; checking WAM, marks, permission codes or supervisor approval (these show as “verify on P&C” instead); and live sync between browser tabs (the SSE endpoint is a heartbeat only).

Data attribution

Course, program and requisite data is sourced from programsandcourses.anu.edu.au, © The Australian National University, scraped via anu-pandc (used with permission) and committed under data/2027/ (see data/2027/SOURCE.md for the exact command and date). A course searched for and not already in that seeded set is fetched live, on demand, from the same site, with a self-identifying User-Agent and a 10-second timeout, and cached from then on.