Eisberg engineering research.
A small set of deep technical papers on the architectural choices behind Eisberg. Posted here when ready, not before. We bias toward depth over volume — one paper a quarter that survives peer review beats four marketing-shaped explainers.
What we are writing.
In review
Agent-native query governance
How identity-bound, audit-trailed agents change the shape of authorization at the planner. The thesis: governance enforced at the planner — not at the consumer — is the only model that survives autonomous agents at scale.
Draft
Compounding intelligence across customers
A privacy-preserving protocol for sharing structural patterns (not data) across tenant boundaries. Includes the anonymity gate, the contributor-count threshold, and the formal proof that no single customer's signal can be reconstructed.
Draft
GPU-native cost modeling
A query-planner cost model that treats GPU memory bandwidth, kernel launch latency, and host-to-device transfer as first-class signals — not exceptions to a CPU baseline. Methodology + measured speedups across analytical, agent micro-query, and federated workloads.
Planned
Tamper-evident lineage at write time
Synchronous open-standard lineage emission with cryptographic chaining at every write boundary. Why post-hoc lineage reconstruction breaks under adversarial conditions, and the design that makes the chain verifiable from the bill alone.
Drafts are reviewed under NDA with select design partners and academic collaborators before public release.
Reviewing research with us?
If you are an academic researcher, principal engineer, or analyst evaluating the agent-native data infrastructure space, we send drafts on request.