I’ve been building something in the background that I think is finally ready for more eyes.
It’s called TrustEdge Labs: an open-source Rust engine for trustable edge workloads: encrypting data at the edge, binding it to provenance, and plugging into hardware-backed key storage when available.
Right now it supports:
- AES-256-GCM encryption with per-chunk manifests and Ed25519 signatures for provenance.
- A universal backend system so different key storage options (software keyring today, YubiKey/TPM/HSM next) can slot in behind the same API.
- A CLI and network mode you can actually run and break on your own machines.
It’s not a product launch (don't use it in production!), just a serious project by someone who cares about privacy, integrity, and verifiable data at the edge.
If you work on IoT, edge AI, or security and have opinions, I’d genuinely love feedback.
Project overview: https://t.co/ZZiDhYg5AJ
Code (MPL 2.0): https://t.co/PQoeDpXiai
When we start to have great ideas about ourselves, God allows greater temptations to come upon us and defeat us, so that we're humbled and seek his help again.
— Saint Isaac the Syrian
Blindjoin is a standalone CoinJoin coordinator and client for Bitcoin signet.
I didn't like the ruckus around @wasabiwallet shutting down coordinators, and the abuse of the Samourai devs, so I wanted to try an idea.
It uses RSA blind signatures (RFC 9474) so the coordinator cryptographically cannot link transaction inputs to outputs. Coordinators are discoverable via @getpubky PKARR DHT and all production traffic flows over Tor hidden services.
Work in progress, signet only, definitely don't trust it with real sats.
Constructive feedback and PRs always welcome!
Maybe someday it will be ready for mainnet. 🙏
https://t.co/Q7ABwko0Qk
First Bitcoin(-related) PR! Woot! 😅
While working on blindjoin, one of my checks kept failing intermittently.
Fable 5 wouldn't touch it (cryptography == hacking, apparently), so I got demoted to Opus 4.8. 😉
Opus tried to wave it off as "random flakiness," but that bugged me. So I dug deeper.
The culprit: the bip322 crate has a sneaky bug. It generates valid ECDSA signatures via sign_simple() but then rejects some of them later in verify_simple() due to a strict "71 | 72" byte length check.
ECDSA signatures aren't fixed-length?! 🤯
They're a pair (r, s):
r = (k*G).x mod n
s = k⁻¹ * (z + r*d) mod n
(k = nonce, G = base point, z = msg hash, d = privkey, n = curve order)
TIL they're usually, not always, 64 bytes (32+32). Sometimes (~1%) r or s is shorter resulting in a valid 70-byte sig after serialization. And sometimes, rarely, both are short.
The signatures were always valid. The verify function just didn't handle all length cases.
Easy fix: Peel off the sighash flag, let the crate's from_der() do proper validation, and add an empty-input guard to prevent panic.
Small change, tiny blast radius. Fingers crossed for review! 🤞
https://t.co/HQaOPdPrKO
My beliefs: Retweets are notifications, not endorsements. Constructive dialogue leads to better outcomes. Bitcoin is hope and economic empowerment for everyone. Every good-faith effort to strengthen the network should be welcomed.
no benchmark will tell you this: LLMs can be /too/ nice
unsurprisingly, in a competitive zero-sum setting, being nice can be bad
i built royale: last agent standing, a br for agents, and ran it 30 times
the nicest model lost hard. the model you least expected, won
🧵:
@davidbombal@ThreatLocker I thought this podcaster was supposed to be legit?
If he's giving guests like this a platform I'm glad I've never listened or watched. And kudos to transcripts so I don't have to actually listen to his guest. 🫡
TL;DR: "we could be totally wrong, but we're not going to consider being wrong because that would mean rethinking and correcting all of the previous wrongness, which would in turn mess everyone up, and that's just silly."
https://t.co/T9qHlqPplr