Three Ways to Give a Blockchain Eyes
Part 3 of “The Oracle Problem” — a plain-language series on the hardest question in Web3.
So far the news has been discouraging. A blockchain can’t see the outside world (Part 1), and AI can’t magically fix that (Part 2). Yet somehow, billions of dollars flow through systems that depend on outside data every day, and mostly they don’t get robbed. How?
The answer isn’t that someone found a way to make a messenger perfectly trustworthy. Nobody has. The answer is cleverer: you spread the trust out so thin that no single liar can do any damage. There are three main families of this idea. Once you see them, you’ll recognize them everywhere in crypto.
Approach 1: The panel of experts
Used by “data feeds” like Chainlink.
Suppose you want to know the temperature and you’re worried one person might lie. The obvious fix: don’t ask one person. Ask thirty independent people, throw out the extremes, and take the middle answer. For a single liar to move that middle, they’d have to corrupt a huge share of your thirty — far harder than fooling one.
That’s exactly how the most widely used crypto oracles work. Many independent operators each report a price. Their numbers get combined into a single agreed value, with outliers ignored, and that is what your app reads. One bad actor can’t budge it.
There’s a subtle catch worth knowing, because it bites beginners: these reports don’t arrive every second. They update when the price moves enough, or when a set amount of time has passed. So a careful app must always check how old the last report is and refuse to act on a stale one. A true price from an hour ago can be as dangerous as a fake one.
Approach 2: Average the marketplace itself
Used by “TWAP” oracles built on exchanges like Uniswap.
The second idea flips the script. Instead of asking outsiders for a price, why not read it straight from a busy on-chain marketplace where people are actually trading? The trouble, as we saw in Part 1, is that the instant price of a marketplace can be violently shoved for a single moment by an attacker with a big enough wallet.
The fix is beautiful in its simplicity: don’t use the instant price — use the average over time. Instead of “what’s the price right now,” ask “what has the average price been over the last thirty minutes?”
Why does this stop the attacker? Shoving the price for one instant is cheap. But to move a thirty-minute average, you’d have to hold the price distorted for many minutes straight — and every second you do, the market fights back and bleeds you money. A one-second heist becomes a thirty-minute siege that costs a fortune. It’s the difference between judging a company by one flickering tick and judging it by its monthly average. The average is boring, and boring is exactly what makes it hard to fake.
(A footnote for the curious: the newest version of Uniswap removed the built-in oracle and moved this job into optional “hooks” — small add-on programs, one of which caps how far the price can move in a single block. Same spirit, sharper tool.)
Approach 3: A quorum of witnesses
Used by “signature quorums,” and increasingly to vouch for AI outputs.
The third idea is the most human. Pick a handful of trusted witnesses. Ask each of them to personally sign off on the same fact — a cryptographic signature that can’t be forged and can’t be denied later. Then make a rule: the fact is only accepted if, say, at least two of the three witnesses signed it.
Now a single compromised witness is harmless. Their lone signature isn’t enough to pass the bar, and because everyone can see exactly who signed what, a bad witness gets caught and removed. This is the pattern of choice when there’s no neat marketplace to average and no crowd to poll — including, increasingly, for vouching that an AI actually produced the output it claims.
The lesson hiding in all three
Look at what these three have in common. None of them makes the messenger honest. None of them removes trust. What they do is refuse to place all their trust in any single point. The panel spreads it across many reporters. The time-average spreads it across many minutes. The quorum spreads it across many witnesses.
That’s the real answer to the oracle problem as it stands today: you can’t eliminate trust, but you can shatter it into so many pieces that betraying the system requires corrupting an impractical number of them at once. And the strongest systems don’t pick just one approach — they combine them, and sound the alarm the moment two methods disagree.
It’s genuinely clever engineering. But it has a ceiling — a kind of problem where averaging and voting and expert panels all quietly fail, no matter how well you build them. That ceiling is where this series ends, and it’s the part that matters most if you’ve ever scanned a QR code to check where your food came from.
Tags: Blockchain, Web3, SmartContracts, DeFi, Fintech
