CAP
Cap
- Confidence
- 23%
Every day a model scans ~200 coins on CoinMarketCap and names the 10 most likely to finish among the week's top-10 gainers. Seven days later every call is graded in public, after costs, against comparable coins and the whole market. No hype, no hidden losses.
Today's 7-day call: CAP · BEAM · SOON · SENT · M · GMX · CHIP · H · PONS · AWE · graded Nov 5, 2026
An experiment, not financial advice. Crypto is volatile and can go to zero.
No confetti. No "up to" claims. No moon talk. Just receipts, graded in public, every call dated.
Today's drop
Confidence is the model's chance that a coin finishes in the week's top 10 gainers of ~200. It is well calibrated: picks scored about 18% on average and hit about 18% of the time in testing. Treat the 10 as one basket: picked coins rise no more often than other coins; the edge comes from the winners running further.
Cap
Beam
SOON
Sentient
MemeCore
GMX
USD.AI
Humanity
Pons
AWE
Picks are model output for research, not recommendations to buy. Most picks don't land in the top 10, and many go down.
The receipts
Every pick gets graded 7 days later. Here's the full record, including the parts that don't flatter us.
Hit rate · finding the week's top-10 gainers
3.2× random
On an average day, 1.6 of 10 picks end up in the real top 10. Random picks would land about 0.5.1
Track record
287
Days of picks since Dec 28, 2025. 271 already graded. Every day is listed in the time machine below.
Real talk
It finds big movers, but it's not a money printer. 44.7% of picks went up over 7 days, versus 47.5% of all tracked coins.2
Best single day
50%
Mar 13, 2026: 5 of 10 picks landed in the top 10. The worst days score zero, and those are in the record too.
Hit rate over time · 14-day rolling average
Loading the full history…
| Month | Days | Hit rate | Random | Went up |
|---|
Time machine
Drag through every graded day. Lime rows made the real top 10. Returns are the coin's 7-day price change.
Loading…
How it works
The 7-day call is the main signal. Each day one model (a logistic regression trained on three years of CoinMarketCap data, frozen, then tested on a year it never saw) turns 18 signals about every coin, plus the market's mood, into one number: the chance the coin finishes in the week's top 10 gainers. The 10 highest are the call. The 3-day call uses a simpler 5-signal formula and is marked experimental: in testing it showed no reliable edge after realistic trading costs.
Momentum
Price change over 1, 3, 7, 14 and 30 days, and how far a coin sits from its 30-day high.
Independence
Beating Bitcoin, and whether the coin moves with its usual crowd or breaks away from it.
Volume
Trading volume against its own normal level and against the coin's size.
Risk
Smaller, more volatile coins move further, both ways. The model knows this and the proof page checks it isn't the whole story.
Overheating
The relative strength index (RSI): coins that rose too fast, too recently, often cool off.
Market mood
Fear & Greed, Bitcoin's share of the market and its weekly trend change how much each signal counts.
Also doing stocks
Stock hit rate vs random
1.3× random
Large-cap stocks move together more than coins do, so the edge is smaller. We'd rather tell you than hide it.
Stock track record
110
Days of stock picks from a universe of ~100 large caps. 49.9% of picks rose over 7 days.
House rules
No streaks, points or celebrations for trading. Regulator studies show that stuff makes people take more risk.
Wins and misses, every single day, with the method written down. No deleting bad weeks.
Every call records each coin's price at pick time, so anyone can check the result themselves.
Every number comes with what random picks or the whole market would have done.
No auto-trading, no "buy now" buttons. It's information. What you do with it is your call.
Anything that moves on this page can be paused, and it respects your reduced-motion setting.
New picks every day. Every pick scored a week later. Bookmark the live dashboard and keep us honest.