Statistics

How many Indonesian words can you make from seven letters

A random rack of seven letters makes a median of at least 23 Indonesian words, but a word using all seven turns up for only a few racks in a hundred. Affixes barely change the first number and do change the second. And our own unscrambler's word list turned out to be film subtitles, about half of it not Indonesian.

27 Sep 2026 · 5 min read

Statistics

How to count fan in Hong Kong mahjong, and why every table scores differently

In Hong Kong mahjong every table agrees that all chows is 1 fan and a full flush 7, but the rarer hands and the payout scale vary. Here is how the fan add up, what a big hand already includes, why tables need 3 fan to win, and how rare each hand is by shape.

23 Sep 2026 · 7 min read

Statistics

How many Chinese characters cover 90% of what you read

About a thousand characters make up 90% of written Chinese, but in film subtitles it is 717, and half of what is said on screen takes just 62. Here are three large text collections counted side by side, where the 3,500 figure comes from, and why 90% of characters is not 90% understood.

22 Sep 2026 · 5 min read

Statistics

The idioms that end a game of 成语接龙

In 成语接龙, the idiom chain game, some idioms leave the next player nothing to say. Here are the dead ends and two-move traps, counted across 29,502 idioms, and how much the house rules, the word list and the script change the game.

21 Sep 2026 · 5 min read

Statistics

AI Alerts Grew 685 Per Cent. They Are 0.43 Per Cent of the Queue.

Three numbers from one study, and 94.1 per cent of these alerts are noise. Each figure supports a different decision, and quoting one alone produces a different story.

12 Sep 2026 · 6 min read

Statistics

One Singapore quarter, two official growth rates, four times apart

In August 2020 the economy had shrunk 41.2% in the second quarter; in October the same quarter was a 13.2% contraction. The basis changed in between, and both bases are still published today — same system, same update date, roughly four times apart. The explanatory note the footnote points to now returns 404.

12 Sep 2026 · 5 min read

Statistics

AI-Disclosing Advisers Grew Headcount Twice as Fast. They Already Were.

Astraeus finds 15 per cent growth against 8 across 6,384 firms. The adoption signal is a line in a regulatory filing, and the report itself says adopters were growing faster beforehand.

7 Sep 2026 · 6 min read

Statistics

The annual Sharpe ratio hides an assumption about your returns

Annualising a monthly Sharpe ratio by multiplying by the square root of twelve is not a unit conversion. It is Andrew Lo's general formula evaluated at zero autocorrelation — and at a monthly autocorrelation of 0.3, unremarkable for anything not marked daily, it overstates the annual figure by roughly a third. The error also runs the other way, which most write-ups miss.

5 Sep 2026 · 7 min read

Statistics

A distress-zone Z-score does not mean a 70% chance of bankruptcy

The Altman Z-score's famous 70% is the share of firms that DID fail which had scored in the distress zone — a sensitivity, measured after the fact. Read as a bankruptcy probability it requires 19.4% of US industrial firms to fail every two years. Our own calculator made this error until this week.

4 Sep 2026 · 6 min read

Statistics

Only Four of 37 AI Infrastructure Rounds Went to New Companies

Of 37 disclosed rounds, 33 went to companies that had raised before, at a median of $275m. The boom is in the companies that already exist rather than in new ones.

4 Sep 2026 · 7 min read