Three mechanics, playable. Built to answer one question: what do puzzles actually exercise, and should the app have them?
Puzzles are not one area, and the honest answer is slightly awkward: most of what people mean by “puzzle” is executive function — and executive function is the one thing Paul’s five working domains deliberately leave out.
Placing them against the pyramid in the domain map:
| Puzzle type | Paul’s level | What it exercises | In the five? |
|---|---|---|---|
| Sudoku, logic problems, ordering | Level 6 | Executive function — reasoning, planning, holding rules in mind | No — parked as “beyond the five” |
| Rebus / “brain twisters” | Level 3, Ex 9 | Set-shifting — you must abandon the literal reading | Domain 2, processing efficiency |
| Which one doesn’t belong | Level 2, Ex 6–7 | Categorization — finding the dimension that sorts them | Domain 2, processing efficiency |
| Crossword, word ladder | Level 3 | Verbal retrieval | Domain 5, recall & communication |
⚠️ Worth putting in front of Paul. The caregiver guide’s six-pillars table describes mental exercise as “Puzzles, learning, memory games — the guided sessions in the app.” Learning and memory games are well covered by the ten games that ship today. Puzzles are the thin word. The closest we get is Say It Straight, which is a word-order exercise rather than a puzzle.
So the guide currently promises the one category the game set was designed to exclude. That is a decision for Paul rather than a defect — the domain map already records Level 6 as left out on purpose. The three options are: build puzzles, soften the word in the guide, or say plainly that puzzles are what a member already does in the newspaper and the app covers what the newspaper can’t.
All three run on the existing engine — content in a spreadsheet column, multiple choice, taps only. No score, no timer, no red. A wrong answer is met with the teaching, never a verdict. Tap through them.
Odd-one-out has a famous failure mode: more than one defensible answer. “Which doesn’t belong — clarinet, oboe, cello, flute?” has a second right answer for anyone who thinks about reeds. The book solves it, and the fix is the hint: naming the dimension collapses the ambiguity to one answer. Every item needs a hint, and the hint is not optional decoration — it is what makes the item answerable. Same class of bug as the Say It Straight anchor appearing twice, and it fails silently: an ambiguous item renders perfectly and just marks a correct answer wrong.
Rebus is the one that can frustrate. When it lands it is the most delightful thing here — the “aha” is the whole reward. When it doesn’t, there is no partial credit and no route in. Keep them classic and warm; a rebus that needs a pun on a brand name is a rebus that ages badly.
Logic problems are where load creeps. Three entities and two clues is a pleasure; four entities and four clues is homework, and the difference is one line of content. The last item here is the interesting one — it teaches that “every Tuesday we meet” doesn’t make today Tuesday, which is a real reasoning habit rather than a trick.
Read-aloud needs clips. A screen with no audio clip doesn’t go silent — it falls through to the device’s own robotic voice, which is what Craig heard on the iPad. Any of these that ship needs its clips generated alongside.