The survey covers the ground where computation, money, and trust are being rebuilt.

The cloud made software easy to ship and easy to lose. Local-first architecture inverts the deal: data lives on the device, sync is a background concern, and the server becomes a peer rather than a landlord. CRDTs and modern sync engines have made this practical; the open question is economic.

We study the engineering — convergence guarantees, partial replication, end-to-end encryption over sync — and the business question underneath it: what does a software company look like when it can't hold your data hostage?

Settlement is becoming a public utility. We track the layer where that happens — consensus design, rollup architectures, data availability, shared sequencing — with the working assumption that most of today's chains are scaffolding for a smaller number of durable systems.

Our protocol work is empirical. We run nodes, replay history, and measure what whitepapers assert: real finality times, real censorship resistance, real cost of verification on commodity hardware.

DeFi is a fifteen-year experiment in removing intermediaries from market structure, run in public, with real money. We treat it as a laboratory: AMM design, intent and solver networks, on-chain credit and under-collateralized lending, stablecoin mechanics, and the slow merger with traditional fixed income.

The interesting questions are microstructural. Who captures the spread when there's no exchange? Where does risk concentrate when there's no clearinghouse? Our market analysis follows flows, not narratives.

The most valuable training data — medical, financial, personal — is exactly the data that can't be pooled. Privacy-preserving machine learning is the set of techniques that resolve the contradiction: federated learning, differential privacy, secure aggregation, and the early, expensive frontier of fully homomorphic encryption.

We benchmark the overhead honestly. Most "private AI" today is a press release; some of it is a working system with a 10× cost multiple that's falling fast. Knowing which is which is the research.

Enclaves let you trust a computation without trusting its operator. That single property — remote attestation — is becoming load-bearing infrastructure for confidential cloud, verifiable AI inference, MEV-resistant block building, and custody systems that no insider can quietly compromise.

We maintain a standing review of the hardware base: SGX's lessons, TDX and SEV-SNP in production, side-channel history, and the open-silicon efforts that could make attestation auditable all the way down.

Software is beginning to hold money. Agents that transact, underwriting models that read unstructured reality, compliance systems that explain themselves — the convergence of machine intelligence and financial infrastructure is where our other five quadrants meet.

The hard problems are not model quality but accountability: who is liable when an agent trades, what an audit trail means when the decision-maker is a weight matrix, and which rails — open or closed — agents will settle on.