Tension: the coiled spring.

The funding rate is the fee traders pay to keep a leveraged bet open. Our validated score reads its sign: crowded longs have historically lagged. But every test we ran confirmed the humbling truth — per-coin direction is close to a coin flip, and anyone claiming otherwise is selling you noise.

So we asked a different question. When the funding fee gets extreme in either direction, the market in that coin is lopsided — one side is paying heavily to stay in the trade. That is a spring under tension: positions are stretched, and stretched positions unwind violently. The hypothesis: the size of funding, sign ignored, predicts the size of coming moves — not the way.

Everyone tries to predict direction, and mostly fails. The size of the move is a different, less-crowded question.

The signal is simple: each day, take every coin’s trailing 7-day funding, drop the sign, and rank. The target: the coin’s realized volatility (how much it actually moved) over the following days. The test discipline is the same one that killed ~17 of our 18 factor candidates:

• Non-overlapping windows only, 2019→2026 — overlapping windows inflate significance, so we refuse them.
• Control for volatility clustering. A coin that moved a lot yesterday tends to move a lot tomorrow — and wild coins attract extreme funding. So we measure only what tension knows beyond the coin’s own trailing volatility (a partial rank correlation).
• A survivorship layer — every test re-run with ~167 delisted coins added back. Backtests that only see the coins that lived are how the industry lies to itself.
• A size/liquidity confound check — is “tension” secretly just “hot small alts”? We re-ran the test controlling for each coin’s dollar volume as well.

On the ~50 established majors at a 7-day horizon, absolute funding predicted forward realized volatility with a rank correlation of about 0.10 (t ≈ 9.6) — and, after stripping out volatility clustering, the residual signal still held at t ≈ 6.0 (stronger at 3 days, fading by 14). It was stable across both halves of the sample.

Then the two killer tests:

• Survivorship halved it — but it survived. With the delisted coins back in, the 7-day t-statistic dropped from ~6.0 to ~3.1. Roughly half the apparent strength was survivorship inflation; the honest half is real. We quote the corrected size everywhere: the most wound-up coins moved roughly 10% more than their own usual over the following days.
• The liquidity confound test came back inverted. Tension barely correlates with liquidity (≈ −0.07), and controlling for both trailing volatility and dollar volume the signal did not weaken (t ≈ 6–8 at 3–7 days). The residual effect was strongest in the most-liquid third of coins — the opposite of a small-cap artifact.

And the property that defines it: direction-agnostic by construction. The most wound-up coins rose about 47% of the time over the following week — a coin flip. Tension knows how hard, never which way.

This result is the survivor of a family of more exciting claims that did not make it:

• “Crowding predicts crashes and drawdowns” — looked strong on surviving coins, then the sign flipped when the delisted coins came back. Dead.
• “Crowding predicts the absolute size of returns” — concentrated in recent years only, and survivorship-inflated. Dead.
• Signed funding as the tension gauge — much weaker than the absolute value. The spring doesn’t care which side is stretched.

Publishing the kills is the point. A “squeeze predictor” would be a better headline; it just isn’t true. What is true is smaller, and we show it at its honest size.

Tension predicts the size of moves; the IOX score ranks expected relative return. Different targets — so tension does not touch the score. It lives in the display layer: a per-coin tension meter on every coin page, a “most wound-up” list in the daily report and the weekly digest, and a bag-tension read in the scanner.

One more reason we built it: for all but two of the coins we score (BTC and ETH), no options market exists — so no implied-volatility gauge of “how much will this move” exists anywhere for them. A funding-based tension read is a rare, honestly-derived substitute, available for every coin with a perpetual market.

A real but modest lean, shown with its whole validation story — including the half that survivorship took away.