Project after project, @CoWSwap trusts us with bigger challenges. Honored to execute BYOS and keep innovating batch auction infrastructure
snapshot.org/#/s:cowgrants.e…
2
6
523
DeFi buybacks leak everything: budget, schedule, every fill, all public. TradFi fixed this in the 80s with dark pools.
We spent the last few days on our first FHE project, a dark pool for protocol buybacks, built on @zama FHEVM. 🧵
1
1
7
2,267
5/ It's a PoC for now: fills are FCFS not pro-rata, treasury solvency isn't onchain, the payment leg is a mock USDT. Next up: pro-rata, encrypted limit prices, real stablecoins.
We're excited to keep building on FHE and take this from PoC to a real product
1
3
146
6/ Live on Sepolia:
🖥️ Demo: confidential-buybacks.vercel…
📜 Code: github.com/bleu/confidential…
Built with @zama FHEVM + @OpenZeppelin ERC-7984 confidential tokens. Our first FHE experiment, but definitely not the last.
#ZamaDeveloperProgram
2
142
3/ ConfidentialBuybacks is that dark pool. The treasury buys straight from holders in sealed epochs, with budget, offers, and fills encrypted onchain.
The whole match happens on encrypted data, no decryption at any point. An encrypted budget draws down first come first served, and once it's gone, later offers silently fill zero.
1
46
4/ Confidential can't mean unaccountable. 5 min after an epoch closes, anyone can trigger public disclosure of the total. Private during execution, public afterwards.
The surprising part of building on FHEVM: it's basically Solidity. The hard stuff (input proofs, threshold KMS, relayer) is infra you inherit, not code you write.
35
A protocol we operate infrastructure for had a 9.5-hour outage last quarter.
The actual fix took a few minutes. The rest of the nine was an alert that sat unread from 00:30 until someone saw it at 09:00. The technical cause: 250GB of old database schemas accumulated over months. A cleanup script existed. It just hadn't been scheduled.
Both root causes converge to one missing branch in the decision tree: who owned this, by when, with what threshold.
Engineering rarely fails at engineering. It fails at ownership boundaries nobody drew. Code shipped without a clear answer to "who pages at 00:30?" is code preserving the absence of that decision.
That's the work a Sprint Zero exists to surface. Before it shows up at 00:30.
2
152
1/ A client on working with us:
"The abyss between a supplier that delivers what you asked for, and one that asks why you want that in the first place, understands what you actually need, and builds that instead."
1
90
3/ We can't work like that. Half the time the most useful thing we do early is push back on what we were hired to build.
The answer might be smaller, a different shape, or upstream of a problem you haven't named yet.
1
52
1/ Shipped and solved aren't the same word. Merging PRs and calling it done is how the real cost gets buried.
1
57
4/ Most of what software costs lives in the gap. Not writing the code. The second pass nobody scoped, the one that turns "it works" into "it works for people who actually use it."
1
34
