Prereq

Programs and Courses is where ANU keeps the rules for every course: what you need first, which semester it runs, what it can’t be counted alongside, and what your degree requires. It keeps all of them as paragraphs. When I planned my own Master of Computing I had to do the rest by hand: work out an order, notice that a course only runs in Semester 2, trace a prerequisite back through two more courses, and keep a list of compulsory courses in my head. Prereq is the version I wanted. It holds the same rules as data, so you can put courses into semesters and it tells you, placement by placement, what won’t work and why.

What good looks like here

It knows which year’s rules apply. Rules change from year to year, and they apply by two different years. A course is judged by the handbook of the year you take it; your degree by the handbook of the year you started. Prereq holds the 2024 to 2027 handbooks and applies each rule by the right one. A 2025 starter is held to the 2025 compulsory courses, not 2026’s. A course placed in 2027 is judged by its 2027 wording. A course placed in a year whose handbook isn’t published yet is judged by the latest one, and the plan says so. Every course page shows its offerings in each year and every change to its requisite wording.

It is honest about what it read. The requisite text on P&C is written by people for people, and some of it has two readings: “COMP6670 or COMP3670 OR (…) and (…) and (…)” can be grouped either way. A tool that picks one silently will one day tell a student they can enrol when they can’t. So the parser refuses anything ambiguous instead of guessing, and a person reads it. Each reading is tied to the exact wording it reads, with the reason written down, and a changed sentence in a new year has no reading until someone gives it one. Across the four years, 38 wordings are hand-read and the rest are read automatically or have no requirement. Every course page says which applies and keeps the handbook’s wording one click away.

It warns and never blocks. The rules come from snapshots of prose, and a student holding a permission code or credit for earlier study knows things this site doesn’t. You can place any course in any semester, twice if you are retaking it; the plan tells you what is wrong and leaves the call to you.

It shows what it doesn’t check. Some requisites ask for a GPA, a permission code or “equivalent” study. Those are marked as caveats rather than implied to pass.

It answers the question P&C can’t. Because each prerequisite is stored as a tree of rows rather than a sentence, every course page can list the courses it leads to. That reverse direction is what you need when deciding what to take first, and the handbook has no way to show it.

Keeping four years side by side showed things no single year does. COMP6250 and COMP8260, compulsory for students who started in 2025, are not offered from 2026. The 2027 handbook still names Machine Learning among the Master of Computing’s specialisations but publishes no page for it, while its Artificial Intelligence specialisation now carries the machine-learning courses. Prereq records the named-but-unpublished specialisation as exactly that.

What is checked, and what is judgement

The automated checks hold the parts that can be stated exactly: the parser’s reading of real handbook sentences in every year, the rule that every refused wording has a person’s reading and every program and specialisation wording its encoded rules, a plan that survives a reload, the right year’s rules applied to each placement and to the degree, COMP8715 taken twice in consecutive semesters, and a crawl of every internal link. Whether each hand-written reading is the right one, and whether the pages are clear to a student at the point of choosing, are judgement calls; the readings are published beside the rules so anyone can dispute them.

What I chose not to build

No accounts: a plan lives at its own address, which you can bookmark or share. Only the Master of Computing’s rules are encoded, because they are the ones I could check against my own degree. Class times and the timetable are out of scope. The handbooks are read once and committed as snapshots (dated on every list page), so the site keeps working when P&C is slow or changes its markup; pnpm scrape refreshes them and pnpm readings lists whatever wording then needs a person.