NDAI and information flow control in DeFi_

This is a two-and-a-half-minute background. One thing Kollan mentioned in his talk is that they don't make every decision in public — if someone wants to do an OTC deal, advertising it to the world has very little promise.[3] Most of what I'll cover is an idea that relates closely to that, called NDAI, from an academic paper by Matt Stephenson and several others.[4]

The basic idea was covered earlier today: if you have private information you want to sell, the buyer can't value it without knowing it — and by revealing it to the buyer, you reduce its value. In the prediction-market case, whoever has to make a decision by proposing it to the market participants undermines the value of the proposal in the first place — exactly the OTC problem.

“The buyer can’t value the information without knowing it. Revealing the information to the buyer, you reduce the value of the information.”

NDAI is easiest to understand with two companies that want to merge. Each has private information; the merger only makes sense in light of the other side's private information. But if the evaluation reveals your information and the deal falls through, you've given up the IP that was the whole business.

So: each party encodes its private information into an agent — the mechanism inside, we leave aside.[5] The agents go into a TEE, a sealed box structured so that exactly one bit of information leaves: deal, or no deal. Inside, the agents negotiate with full knowledge of both sides' information. Once the bit is out, the box's memory is wiped. When there's no deal, there's no leak to the counterparty — or only very limited leakage. It's an interesting theoretical idea. But we can also see it applied in practice.

“There’s only one bit of information which leaves. It says deal or no deal.”

Back in MEV land:[6] in blockchain markets, transactions flow to a block builder — say, us, for this example. As those financial transactions are applied, new arbitrage opportunities appear. But because the transactions haven't finalized and posted publicly, it's not easy for arbitrageurs to capture the opportunity — and those unexecuted transactions shouldn't be revealed publicly, because plenty of adversarial actions can be taken against them.

So there are two ways to let a searcher at the opportunity. One thing we've been running recently is a language where searchers send us many, many transactions in advance, each saying: under these specific conditions, try executing this — it potentially captures an arbitrage.[7] The searcher never sees the flow; their pre-committed bundle only runs if a pool they registered actually moved. The other way: channel the unexecuted transactions into a TEE box that restricts whatever proprietary algorithm is deployed into it. We don't know what the algorithm is; the arbitrageur doesn't see our data outside the box. We send the private flow in, their code generates arbitrage transactions with full information — and because of how the TEE is structured, we know the information can only be used to make those arbitrage transactions.[8]

“We send the private data in, and they generate their arbitrage transactions given the full information. But because of the way the TEE is structured, we know that information can only be used to do arbitrage transactions.”

The first path has been live for a week — very small, and honestly not yet interesting data — and that's where the roughly five-thousand-dollar weekly figure comes from.[9] The sealed-but-sighted path, over the same week, is doing about twice as well. The slide's own caveat applies: two live endpoints in the same week is not a controlled A/B test, and we are not yet capturing most of the opportunity. But same opportunity, different visibility — and the one allowed to look is making roughly double.

That's the idea I wanted to leave you with. The NDAI paper is very much worth reading, and this is proof that it works in practice, at least in some settings. The closing slide says the rest: don't pick a point on the privacy/functionality curve — move the frontier. Privacy here isn't the tax on the trade. It's what creates it.

Every dotted-underline term above, in full, with sources.