Working notes and architecture decision records from R&D across blockchain infrastructure, verifiable AI and post-quantum cryptography.
Most of this started as internal write-ups at Magnus Mage and while building Xyress. Client specifics are stripped out. What is left is the reasoning, which tends to be the part people ask about anyway.
Nothing here is a tutorial. Each note assumes you already know the basics and want the trade-offs.
| ADR | Title | Status |
|---|---|---|
| 001 | Post-quantum migration path for an L1 | Accepted |
| 002 | Treat consensus keys and account keys as separate migrations | Accepted |
| 003 | Anchor AI audit roots on chain, keep records off chain | Accepted |
| Note | Topic |
|---|---|
| crypto-agility.md | Designing transaction and key formats that can change algorithms without a hard fork |
| bft-in-production.md | What breaks first when you run a Tendermint style BFT chain for real |
| verifiable-ai-stack.md | Receipts, TEEs and zkML, and which one a given use case actually needs |
- hybrid-pq-sig: composite Ed25519 + ML-DSA-65 signatures with a height based migration policy
- inference-anchor: Merkle batched receipts for AI inference
ADRs follow the usual shape: context, decision, consequences, alternatives. A superseded ADR stays in place with a pointer to the one that replaced it. Being able to see why we changed our minds is half the value.