Research Partner
ParadigmCheck size: No personal check range or investment authority is published for Storm. Paradigm invests from first check through public markets, but Storm's official remit is frontier data-science and data-engineering research, not a stated personal allocation range.
Storm evaluates decentralized systems through high-resolution data, reproducible measurement, and open tooling. His work spans Ethereum state and history growth, node performance, market-volume accounting, prediction-market visualization, optimization, and forensic analysis. The recurring principle is to replace imprecise narratives with correctly defined metrics, controlled benchmarks, and accessible datasets before deciding where a system's real bottleneck or opportunity lies.
Open data standards, precise definitions, high-resolution evidence, reproducible analysis, instrumented infrastructure, counterintuitive findings supported by data, systems whose true bottleneck can be isolated, prediction markets made legible to a broader audience, and open-source tools that improve engineering decisions.
He publishes no investment pass list and is not presented as a personal check writer. His research implies skepticism toward metrics with double counting, apples-to-oranges comparisons, conclusions based on ambiguous terminology, latency tests that ignore load, scaling claims without hardware constraints, and dashboards or narratives whose source data cannot be reproduced. These are inferred research filters.
Treat a conversation as a technical research review. Define every metric, show raw-data provenance, explain entity resolution and exclusions, and make the analysis reproducible. Separate state, history, access, network, storage, memory, and compute constraints rather than naming one generic bottleneck. For markets, show one-sided flow accounting and adversarial behavior; for infrastructure, benchmark latency, throughput, errors, and failure modes under controlled loads.
Map each business event to its onchain events, identify mirrored maker/taker records, and choose a one-sided measure before aggregating volume or comparing platforms.
Separate state growth, history growth, and state access, then map each to network IO, storage size, memory, and storage IO so optimization targets the actual constraint.
Vary request volume and method under controlled conditions, tracking throughput, latency, error rate, and saturation to reveal performance and failure modes hidden by single-request tests.
Move from raw public data to explicit definitions, documented transformations, high-resolution analysis, visualization, and open tooling so another researcher can reproduce or challenge the conclusion.
Prediction markets become more useful and accessible when their changing landscape can be explored through a browsable, filterable, time-aware map across platforms and topics.
Summing both maker and taker fill events doubles Polymarket volume; comparable prediction-market analysis should use a one-sided maker- or taker-volume metric.
Ethereum history growth is a more immediate scaling bottleneck than state growth, but it is easier to solve and should be analyzed separately through network and storage constraints.
High-resolution evidence shows current state growth can remain within consumer hardware capacity for years; precise separation of state, history, and access is essential to a scientific scaling roadmap.
No podcast appearances are linked yet. Ask Kit to check for interviews that reveal how this investor thinks.
Quote coverage is incomplete. Kit can look for direct comments on markets, founders, and investment criteria.
Storm is a Paradigm Research Partner focused on frontier data science and data engineering. He builds open-source tools for collecting, analyzing, and visualizing data and publishes research on decentralized systems, optimization, forensic market analysis, and prediction markets. He earned a Ph.D. in neuroscience from UC Berkeley using voxelwise modeling to study visual attention, plus a B.S. in bioengineering and B.A. in applied mathematics from Rice University. An earlier official description also identified prior data work at Fei Labs.
RPC infrastructure should be evaluated across controlled workloads and methods because simple unloaded latency misses bottlenecks, degradation, and real performance capacity.