A relative once showed me a sports-prediction service whose sales page listed ten straight wins. The testimonials were glowing. The math took five minutes and told a different story.
Check 1: survivorship arithmetic. If a service texts free picks to 1,024 people, half get "A wins" and half get "B wins," then after ten rounds 1 person has seen ten perfect predictions — by construction. Ten wins in a row is not evidence of skill; it is evidence that you might be the survivor of a binary tree. Ask: how many predictions did they make in total, including the losing ones? If total volume is not published, the streak is unfalsifiable.
Check 2: the take-rate identity. Public betting pools in Japan return a fixed share of the pool (roughly 70–80% depending on the bet type). That means the average participant loses 20–30% of stake, before any subscription fee. A paid tipster must beat the entire market by more than (fee + take-rate) just for you to break even. Plug in real numbers: a 50,000 JPY "premium course" on 10,000 JPY bets needs a sustained edge no published academic study has found.
Check 3: expected value of the refund clause. "Full refund if the prediction misses" sounds risk-free, but write the EV from the seller's side: they keep the fee on wins they did not cause and refund on losses they did not cause. The clause costs them nothing and converts your loss-aversion into sign-ups. Any guarantee whose cost to the issuer is zero carries zero information.
I wrote these three checks down for my relative in plain Japanese. If someone you know is already paying such a service — or being pressured after asking for a refund — there are dedicated consultation desks, and 予想詐欺の相談窓口まとめ explains which desk fits which situation and what records to bring.
The general lesson generalizes to any "alpha-selling" product, crypto signals included: demand the denominator. Anyone selling predictions who will not show you their full track record — losses included — is showing you a marketing funnel, not a model.
Top comments (0)