How a score is made —
and what it doesn’t claim.
IOX ranks coins by the one factor that survived honest out-of-sample validation, expressed as a calibrated probability with its uncertainty. No black box, no “AI predicts price.” Here is exactly how — and where we stop.
Why just one factor? (and why that’s a strength)
Most crypto tools pile on twenty, thirty, fifty “indicators” and call it powerful. It looks impressive. It usually isn’t — most of those indicators don’t actually predict anything. They just make the screen look busy and the tool sound clever.
We do the opposite. We treat every possible signal like a suspect and put it on trial. The test is hard and honest: does this signal really predict what happens next — on data it has never seen — better than a coin flip? And does it tell us something our current signal doesn’t already know? If it can’t prove both, it’s out.
So far, exactly one signal has passed that trial: funding crowding. In plain terms — on crypto futures, when too many people pile into the same bet (say, everyone betting a coin goes up), they have to pay a fee to hold it, and history shows the crowd usually gets punished. We simply measure how crowded each coin is. That’s the signal.
Everything else we tried — open interest, price momentum, volatility, on-chain activity, how thinly a coin trades — failed the trial, for one of three plain reasons:
- It just repeated what funding already said — no new information.
- What looked like an edge was a trick of the data — e.g. the coins that survived to today look like winners, but the ones that quietly died are invisible, so the “edge” was a mirage.
- Once tested honestly, it was no better than luck.
Why is having one trusted signal a strength, not a weakness? Because one number you can trust beats twenty you can’t. Every fake signal you add just buries the real one in noise and hands you false confidence.
We’d rather give you one honest read than a dashboard of pretty lies. Every factor has to earn its place — most don’t, and we publish the ones that failed. That honesty is the product.
Why a focused universe, not 50,000
There are hundreds of thousands of tokens. Almost none are investable. The vast majority are illiquid — you couldn’t buy or sell a real position without moving the price against yourself — and a huge share are abandoned, copy-paste clones, or outright scams, with data too thin or manipulated to mean anything.
Ranking all of them is theatre. A tool that “scores 5,000 coins” isn’t giving you an edge on 5,000 coins — it’s giving you noise dressed as insight. Nobody has a real, repeatable edge on a dead microcap, and pretending otherwise is exactly the hype we built IOX to replace.
So we don’t. A coin earns a score only when you can actually trade it — deep, liquid markets with the derivatives data our engine reads. About 68 coins clear that bar today. But scoring a coin and proving the edge on it are different things — and proof needs years of history. So we split the universe honestly: about 50 coins have the history to validate on, and these are the validated core — the only ones in our public track record and calibration. The other ~18 are newer, liquid listings (SUI, APT, ARB, TAO…) — scored on the same funding signal for coverage, but shown separately and marked “newer · shorter record · not in the validated track.” Same number, less history behind it — never quietly padding the proof. We also tested whether the edge reaches past the liquid universe; across the long tail it decays into noise and disappears. So we stop where the math — and the history — run out.
We’d rather score a focused universe honestly — telling you which coins are proven and which are just covered — than rank 50,000 at random. When a newer coin earns its own track record, it graduates into the validated core, and you’ll see the history that earned it.
The one factor that held up out-of-sample: funding crowding
On perpetual futures, when traders crowd one side of a coin, they pay to hold that position — the funding rate. Crowded positioning tends to mean-revert: when everyone is leaning long and paying up for it, forward returns are typically worse; when the crowd is absent or paying to be short, they’re typically better. So IOX reads funding contrarian.
It’s the one factor that, on our universe, kept its sign across both bull and bear regimes, survived multiple-testing control, and held up when dead coins were added back — directionally real, though not yet statistically significant on its own (survivorship-corrected t ≈ 1.2). The live forward record is what will settle that. Other candidates we tested — simple momentum, low-volatility, on-chain address growth, DeFi TVL, open interest, and illiquidity — did not, and were dropped (see §4).
One honest factor that holds up beats a stack of forty that don’t. As we validate more — each through the same gauntlet, one at a time — they’ll be added with their own weight, and the rest of the method generalises unchanged.
The rigour that keeps it honest
The model is calibrated only on weeks before the one it scores. It never sees the answer it’s predicting.
Test enough factors and one looks great by luck. We control the false-discovery rate, so a factor ships only if it beats that bar.
Every score ships a distribution-free interval. On data the model never trained on, the 80% band covered ~80% of outcomes.
Significance is measured on non-overlapping weeks. Overlapping windows inflate t-stats — a trap we refuse.
We re-ran the validation with delisted/collapsed coins added back. The funding edge held; that’s the honest test most tools skip.
A factor must also produce a tradeable spread after costs — not just a pretty correlation. Several that passed on paper failed here and were dropped.
Does the one factor actually hold up?
A single factor is only worth trusting if it isn’t a fluke of one lucky setup. So we stress-test funding crowding the obvious ways it could break — different market conditions, different time windows, with and without the biggest coins, over different holding periods. Here’s what holds.
The contrarian edge shows up in both calm and volatile markets, and it’s roughly twice as strong when markets are calm. That’s why we surface today’s regime, live (above).
Measure crowding over 3, 7 or 14 days and the ranking comes out about 70–85% the same. The signal isn’t an artifact of one setting.
Remove the five most-traded coins (ETH, BNB, SOL, XRP, DOGE) and the contrarian edge still holds — it isn’t carried by a handful of megacaps.
It points the right way at 3-, 7- and 14-day holds, with 7 days the cleanest read. Consistent across horizons, not cherry-picked.
These checks run on the coins trading today, so — like everything here — they’re directional, not a fresh significance claim (the survivorship caveat in §05 applies). They show the one factor is consistent, not that the edge is large. The live forward record is the real test.
What the number means — and what it doesn’t
It is
- IOX 0–100 — today’s cross-sectional rank, not a probability. 100 = least crowded; IOX 86 means “less crowded than ~86% of the universe today”, not an 86% chance the price rises.
- A calibrated probability (p_beat) — P(this coin beats the average coin in the scored universe over 7 days), shown with its uncertainty band. This benchmark is the peer average — not bitcoin, not dollars. In walk-forward backtests, when we say 54% it landed near 54%; the live reliability curve is accumulating.
- A ~7-day read, rescored daily, with the factor reason visible.
It is not
- A price prediction or a guaranteed signal.
- A promise of profit — any edge is modest and not yet proven; the honest tilt is a few points off 50/50, never 70%.
- Investment advice. It’s research — you decide.
Measured against bitcoin, not dollars
This is how to read a pick’s realised return — we strip out bitcoin’s move so you see the coin’s own performance. It is a separate lens from p_beat above (the probability of beating the peer average) — two different views, not a third benchmark.
Crypto moves as a herd. When bitcoin runs, almost everything runs with it; when it drops, almost everything drops. Picture bitcoin as the tide and each coin as a boat: when the tide comes in, every boat rises — and that tells you nothing about which boat is any good.
So “+10% in dollars” is usually just the tide. We strip the tide out and measure each pick against bitcoin. Try it below — set how much your coin moved and how much bitcoin moved, and watch the dollar gain become the part that’s really yours.
In dollars your coin did +12%. But the whole market (bitcoin) did +4%, so a lot of that gain was just the market rising. On its own, your coin still beat bitcoin.
Two honest notes. Bitcoin is the yardstick here — the market’s tide — not a promise that the signal beats bitcoin; what IOX is built to spot is which coins beat the rest of the pack. And one pick over a few days is still mostly luck — the real proof is the full track record over months, never a single bag.
The honest gate
Factors that fail validation are dropped, not buried. Low-volatility looked great until a survivorship-corrected, multi-regime test inverted the verdict. Plain momentum was noise. On-chain address growth and DeFi TVL came back unproven on a clean test. We show those outcomes because the honesty is the product — it’s what separates this from the “AI predicts 1000x” analysers.
And the real proof isn’t a backtest — it’s the live, forward track record, recorded point-in-time every day and checked against what actually happened. That takes months to become significant. Until then, our survivorship-corrected backtests are directional, not a performance claim — and we’ll say so plainly.