Pair Up (concentration)
Why it earns a place: flashcards show one word at a time. This is the only game where
several words are live in the learner's head at once, so they must discriminate, not just
recall. The interference is the teaching, and it is exactly the muscle that sibling senses
(bangun has six) demand.
What a pair is — the main design choice
- A · word ↔ meaning — kopi ↔ coffee. Shown in the mockup. Simplest, pure recognition.
- B · word ↔ usage — kopi ↔ "Saya minum ___ hangat." Harder, teaches the word in context.
- C · root ↔ form — tidur ↔ tertidur. Teaches morphology; feeds off the affix matrix.
- D · formal ↔ street — menangis ↔ nangis. Feeds off the register work.
Q1. Which pair type ships first — A, B, C, D — or a board that mixes them?
My read: A first (proves the mechanic on data we already have), then C once forms are
flipped visible, since C is the one no other game teaches.
Rules already implied by tonight's fixes
- Never two senses of one lemma on a board — otherwise the same headword appears twice and
the learner cannot tell which meaning it wants.
- Board size scales with device: 12 cards (6 pairs) on a phone, up to 20 on a laptop.
- A pair grades its sense in FSRS on first-attempt success only — luck shouldn't count
as recall. Found clean → Good. Found after a wrong flip → Hard.
Q2. Should a wrong flip cost anything (a timer, a strike), or is the game
purely untimed practice? My read: untimed, no penalty — the memory load is the difficulty,
and timers punish the slow reader rather than the weak recaller.
True or False
Why it earns a place: it is the only format where a learner practises rejecting
something plausible. Every other game asks "which is right?"; this one asks "is this right?",
which is the harder and more transferable judgment — and the one that fails in real conversation.
Where statements come from
- Meaning — "kopi means coffee." False version borrows a sibling's or neighbour's gloss.
- Usage — a real sentence with a real translation; the false version swaps one word.
- Register — "you could say this to your boss." Ties directly into the register program.
- Form — "tertidur is built from tidur." Ties into the affix matrix.
Q3. How near should a false statement be? Far-off falses are free points;
near ones (a sibling sense of the same word) are where the learning is, but they are also where
our data is weakest. My read: near-but-not-sibling to start — same part of speech, different
lemma — and only move to siblings once sense-splitting is trustworthy.
Data risk, worth knowing before either game ships. Both games generate
their content from the same sense data the cards use. A false statement built from a wrong card
reads as true to the learner, and a pair built from two mislabelled senses can have no
correct match at all. Tonight's audit found live examples of exactly that. Neither game should
reach learners before the content-integrity work lands — but both can be built now against
the clean subset.
Both games, shared
- Fed by the same
/api/sessions payload as the cards — no new data path.
- Every answered item is an FSRS review of that sense, same as a card grade.
- One option per lemma, and no question is served unless it has exactly one correct answer.
Q4. Build order. My read: True/False first — it is a day of work, it
runs on data we already have, and it is the better phone game. Pair Up second, once you've picked
its pair type.