James Moore · 22 October 2026 · for James, Dan and Gary. Each lesson becomes a 90-second video or the article, then one short quiz — and an average student who starts in January finishes by late March.
Lessons per unit — the merge
now (999 lessons)after merges (≈750)
How the merges work
A definition and the lesson that uses it become one lesson.
A fact and its "why" become one lesson.
One equation used forwards and backwards becomes one lesson.
Unit 1 barely changes — it is already at the target grain.
Every suggested merge is re-checked against the atom register before it is made. ~750 is the estimate, not a quota.
Two registers — the original grain is never lost
The original register stays as it is, at full granularity.
A v3 register sits beside it and records each combination: this lesson teaches these two or three original atoms.
Questions stay tagged to the original atoms, so results are read at the original grain — a combined lesson whose atoms score differently is a split candidate, and the seam is already on the original register.
Combining genuinely saves teaching time and questions — just not proportionally. Three rearrangements taught together take less time and fewer questions than three separate lessons, but nowhere near a third.
Lessons where students may need the reps anyway (the algebra of rearrangements, for Grade 10s) take the longer quizzes (4–5 questions), and the April bank holds more.
Where the hours go
now: ≈85 hproposed: ≈52 h
Videos: 750 lessons × 90 seconds. A lesson that cannot be taught in 90 seconds gets a more concise lesson, not a longer video.
Quizzes: ~2,250 questions at the agreed per-question timings. The other ~2,300 verified questions go to the April practice bank.
Topic tests keep their current size (47 tests, ~9 questions each) — with one short quiz per lesson, the topic test is where mixed practice lives.
Total ≈ 49.3 h designed; call it ≈52 h with everyday slack. Fast student ≈ 39 h. Weaker student ≈ 65 h.
Video or article — which is quicker?
After the wording pass a lesson's article runs to roughly 180–200 words. At a fluent 200 words a minute that reads in under a minute; the video is a fixed 90 seconds.
So the fast student saves time by reading, the average student breaks even, and the weaker student does better watching — the video cannot be got lost in, and replays cost 90 seconds, not a re-read.
The hours above are priced on the video path — the slower of the two for the students who read well — so the totals are the safe end, not the hopeful end.
Three students against the calendar
The average January starter finishes ~30 March — inside the deadline, with no slack to spare.
The fast student banks a month of extra practice time.
The weaker student has two working routes: start 7 December (and use some of the break), or move to 75–90 minutes a day in January.
The course is ≈3,100 XP — inside Hari's 3,200 XP core target for a January start.
What changes
Every lesson: a 90-second video or the article, then one quiz. One lesson, one video — never split in two.
One quiz, usually 3 questions — more only where the lesson genuinely needs it. The spare ~2,300 verified questions become the April practice bank, tagged by unit, so a practice-exam score report turns straight into gap-fill assignments.
A wording pass over every article under the current writing rules: the absolutes sweep ("never", "cannot", "guaranteed"), US spelling, one symbol per meaning, one notation style.
Dan's findings fixed: the 6 critical errors, the 4 majors (including adding 3d electron configurations), and the named CED gaps.
Dan's Unit-1 rules applied course-wide: mole before "n", no unit-only-on-the-key giveaways, masses to 2 dp, mole answers to 3 sf, four-option questions where a real misconception exists.
Two engine changes: chunked reveal (a meaningful block at a time, not a click per line) and four-option questions.
What stays
The teaching order, and the original register at full granularity — the v3 register records every combination against it.
All 2,212 reviewed figures.
Every verified answer key — a question keeps its cold-solve and blind-read record wherever it moves.
Topic tests and nine unit tests, MCQ and FRQ scored separately.
The review loop: note boxes on every page.
Vote
Who is voting?
1. Adopt the merge map at ~750 lessons?
82 merge groups from the blind review, all following the patterns above.
Each one re-checked against the original register before it is made.
The original register is kept; the v3 register records every combination, and questions stay tagged to the original atoms — so any merge can be split back out on evidence.
Recommendation: adopt it, with the re-check as the gate.
2. Quiz length: three questions as the default?
Three questions a lesson lands the average student at ≈52 h.
A multi-step calculation lesson takes 4–5; a recall lesson takes 2.
Recommendation: 3 as the default, 2–5 by need, the spares banked.
3. Both engine changes: chunked reveal and four-option questions?
Chunked reveal also cuts completion time.
Four options only where a genuine misconception exists for each distractor.
Recommendation: both, in the same pass.
4. The weaker student: December start, or more minutes a day?
At 60 minutes a day from January, a weaker student overruns into April.
Route A: open 7 December and point weaker students at it (plus some break work).
Route B: set their AP block at 75–90 minutes a day from January.
Recommendation: both levers exist — this is really Gary's call on scheduling.