/problems [<course-slug>] [--lessons 1,3] [--due] [--n <count>]

Problem-solving is a distinct skill from recall and is trained separately (see docs/pedagogy.md § Problem practice). Read that and docs/vault-schema.md (problemset schema) first. Vault: <vault>.

Modes

• --due (default when due retries exist): collect all problems whose status line says retry after <date> with date ≤ today — across all courses if no slug given — and run a re-attempt session, interleaved.
• New set (default otherwise): generate a fresh problem set for the given lessons (default: the most recently read lessons without one).

Problem set files

courses/<slug>/problems/NN Problems.md (NN = lesson number), created from templates/problemset.md, frontmatter per vault-schema (type: problemset, course_slug, lesson, problems_total, next_retry). Body per problem:

## P1 · ★★ <short title — the situation, never the technique it needs>
> [!question] <full statement, LaTeX as needed>
> > [!success]- Solution
> > <complete worked solution + grading rubric (what earns which points)>

**Status:** unsolved

Status line values (machine-parsed — keep the exact format): unsolved · failed (retry after YYYY-MM-DD) · solved YYYY-MM-DD.

Generating a set

3–6 problems per lesson, difficulty ★ to ★★★, following the guidance-fading arc: first a worked-example completion (★), middle problems with scaffolding removed (★★), final ones independent — including one transfer problem applying the idea in an unfamiliar context (★★★). Every problem grounded in the lesson's concepts (link them); solutions complete enough to learn from alone; never fabricate numbers a real computation wouldn't give (compute them, with matplotlib/Bash if needed).

Attempt session (terminal)

1. One problem at a time: show the statement, NOT the solution. The user answers with a final result and/or solution sketch; they may also say "hint" (give the smallest useful hint, note it) or "skip".
2. Grade /10 against the rubric, partial credit, elaborative feedback (where the approach diverged, why the correct step works, link concepts).
3. Update the status line: score ≥ 7 → solved YYYY-MM-DD; score < 7 → failed (retry after <today + 3–7 days>) (harder problems get the shorter delay; a second consecutive fail — the status line already said failed before this attempt — gets +3 days and a pointer to re-read the lesson section first). Recompute next_retry in frontmatter as the earliest remaining retry after date in the file (empty if none).
4. A problem that needed a hint but scored ≥ 7 → failed (retry after <today+7>) once more — solved-with-hint is not yet solved.
5. Close out: session summary (scores, what's re-queued for when), append one line to today's study log if it exists, and tell /study users what changed.

Invariants: never reveal a solution before an attempt (or explicit "show me"); never modify <!--SR: lines anywhere; bump updated: on edited files.