How to train

The previous two pieces argued that mental math is not a talent and not a bag of tricks — it is a stack built on fact fluency, pattern recognition, and estimation. If that's true, the training question simplifies. We don't have to invent a new method; we have to do the obvious things in the right order, and avoid a few traps that make most adult attempts stall.

This is the routine our catalog is built around.

What to discard

Before anything else, four practices that feel productive but aren't.

Random tricks from social media give isolated techniques without the structural understanding that makes them transfer. The 11-times trick is impressive on a single example and useless once the digit sum exceeds 9. We don't include trick-only content for this reason.

Generic brain-training apps improve at their own tasks and transfer poorly to arithmetic. If you want to get better at numbers, you have to work with numbers, in shapes that resemble how you'll use them.

Timed drills too early are a particularly common mistake. Timing yourself before you're comfortable doesn't build speed — it builds a stress response. A lot of adults who say they freeze around arithmetic are describing a panic problem on top of an incomplete fluency foundation. Our decks default to no timer until you've got the move clean.

Passive review — flipping through flashcards and reading the answer without genuinely attempting recall — feels like studying and isn't. The learning happens in the retrieval, not the exposure. Every card in /train enforces this with a retrieval-before-reveal interaction: you commit to an answer first, then see whether it was right.

Train fact fluency with retrieval

Some facts deserve to be available instantly because they show up as sub-steps inside everything else. The core inventory: number bonds to 10 and 100, multiplication facts through 10, fraction-decimal-percent anchors, squares 1–32, small powers, and common base-10 logs.

Concretely, that means decks like number-bonds-10, number-bonds-100, x-tables-2-9, squares-1-32, powers-of-2, powers-of-3-5-7, log10-2-3-7, and fraction-decimal-anchors.

The goal of fluency training is not memorization for its own sake. It is to make these facts cheap enough that they stop occupying attention. Once 7 × 8 surfaces in 200 milliseconds rather than two seconds, every problem that contains it becomes lighter, and your working memory gets to spend itself on the parts that actually require thought.

Train strategies in families

The second pillar is pattern recognition, and the way to build it is to group problems by structure rather than learning isolated tricks.

compensation-add is a family: 49 + 38, 97 + 46, 198 + 75 — twenty problems built around the same round-up-and-correct move. After enough exposure, you stop seeing them as separate problems. The same logic drives compensation-subtract, counting-up, near-round-mul, halve-and-double, and benchmark-percentages.

Train one family in a session, not a mix. Mixing all strategies randomly during early practice produces less fluency than spending a focused block on one structure. Once a family feels automatic, mixing is fine — it's even useful — but it isn't the way in.

Train estimation as a primary skill

Estimation is the third pillar, and it should be trained on its own rather than treated as a watered-down version of "real" arithmetic. The point isn't a less-precise answer. It's a different output: a range, committed to before any calculation.

Practice the move directly. x-near-anchors and div-near-anchors train the multiplicative shape of "ballpark first, refine if needed." pct-from-anchors trains it for percentages. order-of-magnitude and compound-scaling push into the territory where the error you care about is "off by 10x," not "off by 5%." sanity-check is the capstone — every card is an output you have to either accept or flag as wrong, with the bias deliberately tilted toward correct answers because complacency is the actual failure mode in real life.

The standard that matters is not precision. It is whether you can call out a number that doesn't make sense, fast.

Order matters: accuracy, comfort, speed, pressure

The sequence is accuracy first, then comfort, then speed, then pressure — in that order. Most adults shortcut it by adding pressure early, and the stress response is real: working memory shrinks, facts become harder to access, you prove to yourself you can't do it, you eventually believe it.

Don't practice panic and call it fluency. Once a move feels comfortable, moderate time pressure becomes useful, and the deck UI exposes a timer when you ask for one. Most people don't need competitive timing.

A minimal routine

Ten minutes is enough.

• 2 minutes of fact retrieval — pick one fluency deck.
• 3 minutes of one strategy family — pick one and stay there.
• 3 minutes of estimation — usually x-near-anchors or sanity-check.
• 2 minutes on a real-life check — the bill, the dose, the spreadsheet cell.

Daily beats weekly. The interleaving across pillars keeps each one warm without overloading any single muscle.

What you're trying to achieve

The milestone is not a multiplication trick at a party. It is quieter: wrong answers start to feel wrong before you have verified them. That feeling — number sense as a low-level filter running in the background — is what this skill produces, and it's why we train in this specific order.

This is a product-context distillation. The long-form essay lives on mingjerlee.com →.