Say three different people tell you the same stock is worth owning right now. One is a wallet that has been quietly accumulating on-chain for two weeks. One is a company's CFO filing a Form 4 to report she bought shares. One is a member of Congress disclosing a purchase on their periodic transaction report. Each of them is telling you something, and the interesting question is not whether any single one is right. It is whether the three of them agreeing means more than the sum of the parts.
It does, and the reason is worth being precise about, because it is easy to get wrong. The value is not that you have three signals. The value is that you have three signals that did not talk to each other. A whale accumulating ETH does not know or care what a senator's blind-ish trust is doing. A biotech insider buying her own stock ahead of trial data is reacting to information a congressperson has no access to, and vice versa. Different motives, different information sets, different timing. When streams like that independently point at the same name, the odds that all three are noise at once get small fast.
Independence is the entire point
Here is the trap. If you stack ten signals that are all really the same signal wearing different hats, you have not built confidence, you have built an echo. Momentum, RSI, and rate-of-change on the same asset over the same window are basically one measurement described three ways. When they agree, of course they agree. They are correlated by construction. Treating that as three confirmations is how people talk themselves into positions that were always just one idea.
The math backs the intuition. If you have three independent signals that are each right, say, 60 percent of the time, the chance that all three fire falsely on the same name in the same window is roughly the product of them being wrong, which is well under ten percent. But that multiplication only holds when the errors are uncorrelated. The moment the signals share a common cause, the errors correlate, the product collapses, and you are back to trusting one thing. So the useful question for any signal you add to a stack is not "is this predictive" but "is this predictive for reasons the others don't already capture."
Whales, insiders, and Congress clear that bar in a way that most technical indicators never will. They come from genuinely separate corners of the world:
- On-chain whale flow reflects positioning by large holders and is visible the moment a transaction confirms.
- Form 4 insider buys reflect what officers and directors think about their own company, filed to the SEC.
- Congressional disclosures reflect trades by people with a specific and much-argued-about seat near policy and legislation.
Three different motives, three different edges. That is what makes their agreement mean something.
Conflicting signals should cancel, not average
The flip side matters just as much and gets handled badly all the time. When two independent signals disagree, the instinct is to average them and take a middle position. That is usually wrong. If a whale is accumulating hard while insiders are dumping into the same name, you do not have a lukewarm buy. You have two informed parties who see the situation differently, and the honest read is that you do not know which one is right. The correct output there is closer to zero conviction than to half conviction.
Averaging quietly assumes the signals are measuring the same underlying truth with some noise, so splitting the difference lands you near it. But independent signals with different information are not noisy readings of one number. They can genuinely point in opposite directions because they know different things. When they conflict, the disagreement is itself the information, and it should suppress the position rather than water it down. At Blockcircle we lean hard on this. Agreement compounds conviction, conflict cancels it, and a single loud signal on its own is treated as a lead to watch, not a trade to size up.
The lag problem nobody warns you about
Now the part that makes this harder than it sounds on paper. These three streams do not arrive at the same time, and the gaps are large.
On-chain data is effectively instant. The transaction confirms, it is on the ledger, you can see it. Form 4 filings are quick too, since insiders generally have to report within two business days of the trade. Congressional disclosures are the slow one. Under the STOCK Act the reporting window runs up to around 30 to 45 days after a transaction, and in practice plenty of filings land right at the edge of that window or drift past it.
So when you see a congressional buy today, the trade itself may be over a month old. If you naively line up "these all happened recently" by filing date, you are comparing a fresh on-chain move against a decision someone made five weeks ago. The prices are different, the setup may have changed, and the confluence you think you see is partly an artifact of when the paperwork showed up. The fix is to align on transaction date wherever you can back it out, not on disclosure date, and to hold an honest window. A congressional buy from six weeks ago plus a whale accumulating this week can still be real confluence, but only if you are stacking them on a timeline that accounts for the lag instead of pretending everything is simultaneous.
Lining up the tickers is its own headache
The last practical wall is boring and constant, which is matching a name across datasets that were never designed to agree. On-chain you are dealing with token contract addresses and symbols that collide, since half a dozen tokens can share a ticker and one of them is the scam. Form 4 filings key off CIK numbers and issuer names. Congressional disclosures are frequently free text typed by a staffer, so you get "Apple", "Apple Inc.", "Apple Inc", and the occasional "AAPL" for the same holding, sometimes with the asset type mislabeled.
Resolving all of that to one canonical entity, mapping tokens to their real contract, tying tickers to CIKs, normalizing the messy human-entered names, is unglamorous work, and it is where a lot of the actual edge lives. If your entity resolution is sloppy you will either miss real confluence because the names did not match, or invent fake confluence because two different companies got merged. Get the plumbing right and the interesting cases surface on their own.
None of this is a magic button. Confluence narrows the field and tells you where to look harder, and you still have to size the position, respect the lag, and accept that sometimes three independent parties are all early or all wrong together. But three streams that don't talk to each other landing on the same name is a very different thing from one indicator shouting, and it is worth building the pipes to notice when it happens.