“Follow the money.” If candle patterns are what beginners stare at, flow is what the next tier swears by: the funding rate, open interest, spot volume versus futures volume. The pitch is seductive — stop looking at the price, look at who is positioned and how they’re paying for it, and you’ll see the next move before it happens.
It’s a good instinct, and a testable one. So we tested it the way we test everything before it gets near the bot: against years of real data, with the answer locked in before we looked.
What we did
We pulled the real series the playbook is built on: every 8-hour funding payment since 2020, and — for the “leverage” claims — 5.75 years of actual open interest plus spot and futures order flow (12,594 four-hour bars). For each signal we wrote a precise definition, locked it, found every occurrence, and measured what price did 1 to 14 days later against what Bitcoin does anyway. A bearish signal only counts if price did worse than the market’s normal drift; a bullish one only counts if it did better.
Method, briefly
Definitions pre-registered, no tuning after the fact. Edges are stated versus Bitcoin’s baseline drift over the same horizon. Every survivor had to hold up across four time-ordered slices of history (not just the easy years) and on a non-overlapping sub-sample, so one lucky bull run can’t carry it.What we found
| The signal | What people say | What the data said | Verdict |
|---|---|---|---|
| Deeply negative funding | “A dump is coming” | Price drifted up ~+1.8pp / 7d | REAL — BUT BACKWARDS |
| Funding dropping / flipping negative | “Shorts are loading up” | No edge beyond the level itself | NOISE |
| Rising open interest, no spot buying | “Leveraged rally — it’ll unwind” | Didn’t fade: ~+0.1pp, a coin-flip | NOISE |
| Low-volume / weekend pump | “Fake — it reverts” | Reverted by ~0.3–0.5pp | TOO SMALL TO TRADE |
| High-volume move | “Real — it continues” | Continued by ~0.3pp | TOO SMALL TO TRADE |
| Risk-off macro (VIX & dollar up) | “Bitcoin sells off” | −1.4pp / 7d — but only in stress years | REAL — ONLY IN A CRISIS |
Start with the big one, because it’s the single most-repeated flow signal in crypto: negative funding does not mean a dump — it mildly means the opposite. When funding goes deeply negative, the crowd is already paying to be short; that’s fuel for a squeeze up, not down. Price drifted about +1.8pp above its normal pace over the next week. The effect is real and consistent — but small, and the “funding is dropping, shorts are coming” version adds nothing on top: it’s just a slower way of noticing funding is low.
Then the one we most wanted to be true: “spot is real money, leverage unwinds.” The idea is that a rally carried by futures — open interest climbing while spot buyers sit out — is hollow and snaps back. We had the actual money to check it: real open interest, real spot and futures order flow, 5.75 years of it. Among strong rallies, the “leverage-led, no spot behind it” ones faded by about +0.1pp — which is to say, not at all. A rally with no spot buying behind it was about as likely to keep going as to roll over. The story is great. The data shrugged.
The rest is a tour of faint and fragile. Low-volume and weekend pumps do give a little back, and volume-backed moves do carry a little further — but by a few tenths of a percent, gone the moment you pay the spread. Risk-off macro is the one with real teeth: when the VIX and the dollar spike together, Bitcoin underperformed by ~1.4pp. The catch is that nearly all of it came from a couple of crisis windows; in calmer years it vanished. A signal you can only trust during a crash is hard to lean on.
The honest conclusion
We’d have happily wired any of these into the bot — that’s the whole reason we ran them. But “real but backwards,” “real but tiny,” and “real but only in a crisis” are three flavors of the same answer: not a usable edge. The instinct behind flow-trading — think about who’s positioned and what will force their hand — is genuinely good. The specific signals people repeat just don’t beat a boring trend rule that already sits out the chop. So the bot stays boring, and the negative result gets published. That’s the point.