Skip to main content

Technical Innovation Summary

SENEX is built on a set of named inventions. Several are live in v1.0-alpha; others are named research directions on the roadmap. We describe what each one does — the underlying constructions are proprietary.

Live in v1.0-alpha​

  1. Proof of Intel — a contribution's Proof of Data and Proof of Computation fuse into a single verifiable record that decides each participant's reward share. This is the core of the intelligence economy: value is attributed to who actually moved the result, not to who showed up.
  2. Neurolinks — governed, revocable grants over a participant's data paths. They let an agent contribute the impact of its data without ever exposing the raw data or even the raw locator. Approval is explicit; revocation is immediate.
  3. Block-style verification — every participating node independently recomputes each iteration and seals a hash-chained record before anything settles. A node that submits a false result is detected by the quorum and slashed. Verification is by full replication, so extra identities cannot out-vote the truth.
  4. Coordination surplus with fair side payments — when participants coordinate (the canonical two-traveler routing example), the network reaches an outcome that beats the self-interested equilibrium, and the measured surplus funds exact side payments so that no participant ends up worse off than acting alone.
  5. Pull-based, governed data sharing — contextual contribution under owner control, never always-on data mining.

Named research directions (roadmap)​

  1. PCIT — a mechanism to verify that an AI computation was performed correctly without re-executing it in full, so verification stays cheap as the network grows.
  2. CWPS — sybil resistance for an open network: making it economically irrational to contribute false data or spin up throwaway identities, without trusted hardware.
  3. BRSR — a bounded-regret approach to reward attribution that stays fair even when perfect attribution is provably impossible.
  4. RSC-TGV — the "knowing over learning" thesis: a path to capability that composes existing understanding across agents rather than relying solely on fresh training.
  5. Formal privacy stack — client-side differential privacy, secure multi-party computation, and zero-knowledge proofs, extending today's local-first sovereignty to the statistical traces of contribution itself.

The capability of each invention is stated plainly; the constructions that make them work are not published.