Mental Math Training

Mental math training is a short arithmetic practice tool for improving speed, accuracy, and number sense without relying on paper or a calculator. It generates addition, subtraction, multiplication, and division problems at different difficulty levels, then measures how quickly and correctly each answer is entered. The purpose is not only to get a score, but to build fluency with common operations, reduce hesitation, and reveal which problem types cause mistakes. Common practice contexts include students, exam warmups, classroom drills, adults who want a quick brain exercise, and anyone building everyday calculation confidence. The results are session-based, so they should be treated as practice feedback rather than a formal grade or long-term assessment.

Each session is a sequence of independently generated problems. Operands are drawn with Math.random(), which in V8 is the xorshift128+ pseudo-random generator and in SpiderMonkey is XorShift128+ as well — fast and statistically uniform, but explicitly NOT cryptographically secure. For an educational drill that is the right choice; if a future version ever ties practice scores to prizes, the operand draw should switch to crypto.getRandomValues(new Uint32Array(1)) so the sequence cannot be replayed by reading the page's seed. Difficulty levels gate operand ranges (easy 1–9, medium 10–99, hard 100–999) and division problems are generated quotient-first then multiplied to guarantee an integer answer, avoiding the awkward case where 7 ÷ 3 would force a decimal entry. Latency on each answer is measured with performance.now(), a DOMHighResTimeStamp with sub-millisecond resolution (5 µs in Chrome, 1 ms in cross-origin isolated contexts under Spectre mitigations), captured at keydown of the submit and subtracted from the timestamp of the problem render. The final score is the simple proportion correct/total, while average time is the arithmetic mean of per-question latencies and total time is their sum. State lives in React component state for the duration of the session, with no persistence layer — closing the tab loses the result by design, since the tool is a warm-up rather than a longitudinal tracker.

• Math.random() in V8/SpiderMonkey is xorshift128+ — fast, well-distributed, NOT cryptographically secure; fine for practice but not for prize draws
• crypto.getRandomValues() is the CSPRNG (Web Crypto / RFC 4086) — use it instead of Math.random() if randomness must be unpredictable to the player
• performance.now() returns a DOMHighResTimeStamp with µs precision in Chrome, clamped to 1 ms in non-isolated contexts as a Spectre mitigation (specified by W3C High Resolution Time)
• Division problems are built quotient-first (divisor × quotient = dividend) to guarantee an integer answer without trailing decimals
• Operand ranges per difficulty: easy [1,9], medium [10,99], hard [100,999]; the ranges are uniform, so carry-heavy combinations like 7+8 appear no more often than 1+2
• Accuracy = correctCount / totalCount; average time = Σ latency_i / n; both are session-scoped — closing the tab clears state because there is no storage layer
• For long-term practice, log scores manually or wire a localStorage entry keyed by date — the page itself keeps no history across reloads