Roadmap

The Implementation Plan, the PR Roadmap and the Parallelization Strategy were written before any of this existed. They drifted, in two ways worth naming so this chapter does not drift the same way.

They name files. Every migration filename in the PR roadmap is now wrong, because timestamps moved when the work was actually done. This chapter names models and controllers instead; those survive a rebase.

And they know only two states, planned and done. Most of what is left is neither: the code is written, it sits in a pull request, and it is waiting for a person. An entry that says "planned" for three months while a branch rots is worse than no entry, because it hides the real blockage. Every entry below therefore ends with what it waits on, and whether that is code or a decision.

The shape of what is left

Everything below is one chain. Points have to be writable before anything can be published, and nothing can be published before a student has a page to read it on. The two halves are split that way here: what puts data in, and what lets someone see it.

graph TD
    I["Merge the Müsli integration<br/>PR #1211"]

    subgraph Writing["Writing — teachers and tutors"]
        direction TB
        TUT["Points for tutorial groups<br/>PR #1150"]
        TALK["Grades for talks<br/>PR #1196"]
        EXAM["Points and grades for exams<br/>nothing written"]
        ACH["Marking achievements<br/>nothing written"]
    end

    subgraph Reading["Reading — students"]
        direction TB
        RES["What a student sees<br/>points · grade · submission · achievements<br/>nothing written"]
        DASH["Dashboards<br/>no specification"]
    end

    I --> TUT --> TALK
    I --> EXAM
    I --> ACH
    TUT --> RES
    EXAM --> RES
    ACH --> RES
    RES --> DASH

    classDef inpr fill:#dbeafe,stroke:#1d4ed8,color:#1e3a5f
    classDef todo fill:#fee2e2,stroke:#b91c1c,color:#5f1d1d
    classDef spec fill:#fef3c7,stroke:#b45309,color:#5f3d0d
    class I,TUT,TALK inpr
    class EXAM,ACH,RES todo
    class DASH spec

Writing — teachers and tutors

Nothing in the merged tree writes a point. Every point in the system comes from lib/demo. Grade schemes, performance records and the whole eligibility chain run, but only ever on seeded data. Until that is fixed, everything downstream is built on a floor nobody has stood on.

1 · Merge the Müsli integration

The five former slices in one branch, with the Playwright suite for the assessment area and its own round of review. Every other entry either builds on it or conflicts with it, so nothing below can start until it lands.

2 · Points for tutorial groups

The read-only point grid, the grade scheme pipeline, and the performance and eligibility chain are all fed by hand-entered points that no interface can produce. This is that interface — for tutorial groups, and nothing else. Talks, exams and achievements each need their own path; entries 3 to 5 are those paths.

Open since May. Before a review it needs a duplicate note migration dropped — the integration branch folded that column into the create-table migration — and a rolled-back schema version fixed. CI has still never run for it: the test workflow fires for pull requests based on main or next only, and this one targets the integration branch.

3 · Grades for talks

The same surface for talks. Grades can otherwise be written for a whole cohort by applying a scheme, and in no other way — there is no path for a single correction.

4 · Points and grades for exams

An exam can be registered for, scheduled, and graded by applying a scheme to a whole cohort. A single result cannot be entered or corrected by hand at all. This is the exam-shaped version of what entries 2 and 3 build: points per participation, the grade beside them, and the correction path the scheme-wide application does not offer.

Publishing belongs here too, as two buttons rather than a project of its own. results_published_at is written by nothing but factories today, and Exam#status is its only reader — it decides when an exam counts as graded, and so when a student may see the result. Tutorial points need no such switch.

5 · Marking achievements

Achievements exist, the marking view exists, and the service that feeds them into performance records exists. Nothing marks one except a seed, so the eligibility rules that count achievements count only fixtures.

Reading — students

The assessment surface is teacher-only from end to end. AssessmentAbility grants everything through the lecture's edit right, and no route reaches a student.

6 · What a student sees

A student cannot see their own points, grade, submission state or achievements anywhere, and all four belong on the same page. Nothing here can start before something writes them, so this waits on entries 2 to 5. What it needs is the student's own view of a participation, and an ability that grants it to the participant rather than to the lecture's staff. For achievements that includes progress: what is earned, and what is still open.

It is also where the submission page has to be rebuilt rather than extended. Today it answers one question — did I hand something in — and results have no place on it. A student looking for a point, a grade and a submission state will look in one place, and that place is the wrong shape for two of the three.

7 · Dashboards

The largest untouched block, and the only one where no code exists at all. Worth re-specifying before starting: the chapters describing it were written before the assessment area existed and assume screens that were never built.

Then — what is left over from registration

Exam registration, second pass

Making an exam the fourth thing a student can register for — next to tutorials, talks and cohorts — left four screens half-fitting. The defects are fixed. What remains is design, and it was deliberately kept out of the integration branch so that it would not be reviewed inside a very large diff:

• what the "not assigned to a group yet" notice should say when an exam is one of the things a student holds, given that an exam involves no allocation at all
• whether that block should know about the campaigns, so it can speak per kind instead of once for everything
• whether the lecture roster should be a prerequisite for registering — today it is not, deliberately, and registration is decoupled from content access
• the tutorial-flavoured styling an exam tile inherits

It waits on entry 1, then a design call.

Continuous

CI health

The two named failures are gone. The end-to-end spec that had been red on next was a blocked PDF loader, replaced by a separate rasterizer in PR #1216; the strict-mode violation that followed — a second login button whose accessible name contains the first one's — was PR #1218. Both are merged.

What remains is the intermittent one: the Playwright suite fails roughly two runs in eleven, always the same shape, a test timing out with a navigation in flight. Measured and excluded: browser memory, server latency, asset serving, recompilation after an edit, background jobs, disk, CPU, and unhandled dialogs. Not reproducible on demand. Recorded here so the next person does not start the search from the beginning.

A permanently red pipeline hides real regressions, so the rule stands: fix it or skip it, and say which in the pull request.

The book

The chapters were written ahead of the code and have drifted. The known cases, each small on its own:

• five controllers documented as if they existed; two live only on unmerged branches, three nowhere
• a whole section of student-facing views that no user can reach
• the roster namespace written in the singular throughout, while the models use the plural
• the policy engine described as a case statement, where the code dispatches to handler objects — and the service that actually decides eligibility has no chapter
• the evaluator's result described with fourteen fields where it has five, and without the deferred status the whole mechanism exists for
• duplicated sections and one file that is a heading and nothing else

None of it is hard. All of it costs a reader an hour of confusion, and some of it will send them down the wrong path.