Just shipped ClauseRoot for the @GenLayer Agent Tank. Most upgrade systems answer one question: who is allowed to upgrade the protocol? ClauseRoot adds the question that matters just as much: should this proposed code be allowed to become the protocol at all? A proposer submits the exact contract source, pinned to an immutable GitHub commit. The ClauseRoot Governor verifies that evidence, then GenLayer validators evaluate the proposed code against fixed constitutional rules covering fee limits, user custody, withdrawal rights, and upgrade authority. The verdict is stored on-chain. In the live deployment, V2 was APPROVED and executed. Then I submitted an intentionally hostile V3 that tried to raise the protocol fee to 10% and introduce a Governor bypass. GenLayer consensus rejected it for `C1_FEE_CEILING` and `C4_GOVERNOR_BYPASS`. The important part: the rejected code never became the protocol. The live target stayed on V2, with its state preserved. This is not a frontend pretending to make governance decisions. The frontend submits and reads state. The Governor and GenLayer consensus determine the outcome. Demo: youtu.be/MP4PPf4pc0U Live app: clauseroot.vercel.app GitHub: github.com/Habuskid/clausero…

What does risk management look like when an asset keeps trading, but its primary price-discovery venue is closed? That question became ClosedBell, my submission to the AI Trading Desk track, specifically Decision Stress Testing, for the @Bitget_AI Hackathon Season 2. rTokens present an interesting market structure problem. A tokenized US stock can remain tradable outside the regular trading hours of its underlying stock. The ability to trade continues, but the information environment changes. The last regular-market price becomes stale. New information can emerge before the underlying market reopens. Liquidity can change. And a position that looks reasonable in isolation can create significant concentration risk when considered as part of an existing portfolio. Most trading interfaces are good at answering one question: “What is trading right now?” I wanted ClosedBell to answer a different one: “What does this decision change?” Instead of trying to predict where Nvidia will open tomorrow, or reducing a complex decision to a simple BUY/SELL signal, ClosedBell begins with the trade you are actually considering. Suppose I want to add 5,000 USDT of rNVDA while Nvidia’s home market is closed. I give ClosedBell the proposed position, my existing portfolio, and the risk limits I am willing to accept. ClosedBell then stress-tests the portfolio with and without that trade. It shows what changes, whether the new position remains within my limits, and which condition becomes the constraint. The result can be: PROCEED RESIZE HEDGE WAIT INSUFFICIENT EVIDENCE But the outcome itself is only part of the product. I wanted to understand what changed, which limit actually mattered, what smaller position could fit when the original one did not, and what information the decision was based on. That last part is important. A recommendation without context is difficult to trust, especially when the underlying market is closed and the information environment is incomplete. So ClosedBell keeps the evidence, timestamps, and assumptions attached to the decision. It also creates a decision receipt that can be inspected later, preserving not just the outcome, but the information available when that decision was evaluated. There is an important boundary here: ClosedBell does not execute the trade, control a wallet, or pretend to know tomorrow’s price. The final decision remains with the trader. The thesis behind ClosedBell is straightforward: When continuous trading meets discontinuous price discovery, the useful question is not simply whether you can still trade. It is whether you understand the portfolio consequence of doing so. That is what I built ClosedBell to examine. Bitget AI Hackathon Season 2 Track: AI Trading Desk Focus: Decision Stress Testing Live product: closedbell.vercel.app Source code: github.com/Habuskid/closedbe… Full demo below.