Want to join our perps closed alpha later this year? DM.

1/ This is unfolding in real time. I just discovered that this is a sophisticated automated attack. Here's how 🧵

1/ No perfect perp DEX, only tradeoffs Topic 2: Mark Price Design Ideas

You're 5 steps from quoting live Solana flow: 1. Get a maker_id 2. Dashboard: quoting key, enable markets, deposit 3. Add the SDK to your project 4. Initialize the client + quoter 5. Submit a quote docs.flintlabs.dev/overview/…

Canada Demo Day - Colosseum Crypto Worlds Fair nitter.siddz.com/i/broadcasts/1OxwbnMQB…

ADL: Surprises and Fairness A year on from 10/10, I wanted to write a little about how @flint_trade_ will be thinking about ADL, as that was the dominating mechanic on that fateful day. First, a quick recap. ADL (Auto-Deleveraging) generally refers to the venue forcibly closing profitable positions, or haircutting positive PNL, to cover ‘bad debt’. Bad debt is what the market accrues when a user’s losses exceed their collateral and their equity goes negative before they can be liquidated. The market is basically insolvent if everyone was ‘closed out’ at that moment in time. So ADL is used to fix things when all else has failed (liquidations, backstop takeovers, the insurance fund). The bad debt must be covered or the exchange and everyone’s capital would be at risk. Sad but understandable. So the real question is what mechanic and decision framework to use for that? What objective function to focus on and maximize? At a high level, the biggest themes within the ADL problem space are ‘surprises’ and ‘fairness’. ‘Surprises’ are not something desired in a risk engine. You want predictability. You want to know the terms and conditions of the venue you are trading on: how it will act and react given market conditions and when certain stressors are triggered. While most exchanges publish some general guidelines of how/when ADL will be triggered, some of the rules are quite opaque and they usually reserve some ability to put their thumb on the scale, leaving some traders blindsided by how much of their equity is seized, and under what rules. On most venues, you can’t check any of this. You find out who was picked and why after the fact, if at all. On the other hand, ‘fairness’ is basically trying to find the right balance between sharing the load across all market participants (socialization) versus concentration on the users/positions that introduced the highest amount of risk on the venue (leverage targeted). From different perspectives, both can be seen as fair. Socialization is ‘nice’ because on a per-user basis the hit is small. Leverage targeting is ‘nice’ because the cost lands on those who brought the most risk. At @flint_trade_, amongst other things, we are aiming for alignment between the venue and its traders, and for verifiability, which has measurable effects on our ADL design. While at the time of writing we (mainly the resident quant @davethereiss) are aggressively stress-testing the parameters of the final design, we feel strongly about the following: - Balancing between socialization and risk-concentrated users. - Verifiable on-chain targeting: the rules that decide who can be deleveraged, and the limit on how much, are enforced on-chain. The idea will be that traders know what they are getting themselves into and will be able to plan accordingly if they have certain requirements. Additionally, while users introducing more risk to the exchange will share a larger load of covering the bad debt, they will not be immediately ‘wiped out’ when ADL is triggered. Positions are closed at the bankruptcy price of the account that failed, and only positions in profit at that price are eligible, so a trader who gets ADL’d still walks away with a profit on what was closed. Putting elaborate financial rules on chain is not for the faint of heart. It is far harder from the code and design perspective, but it is worth it for the seriousness of the commitment. The general idea is that every eligible position gets a score based on its PNL% and its leverage. A threshold sets the minimum score needed for ADL. It starts high, so only the highest-scoring positions qualify, and decays over time, faster if the bad debt isn’t being cleared. Each hit is capped to a portion of the position and followed by a cooldown, so the load spreads instead of landing on one trader. ADL instructions will be submitted by incentivized keepers, but the program checks every one of them on-chain: a keeper can only touch positions that clear the threshold. While alignment is a broad term, it is something we take seriously and it comes up in every design discussion. As we get closer to launch, I’ll continue to share more thoughts about the various mechanics underpinning the Flint Perps system. Of course, formal specifications for all this will be published as articles and eventually open-sourced code. DG