Skip to content

RESEARCH

A direction, not an unearned claim.

Superintelligence is the long-term ambition, not a capability the protocol claims to have. The near-term objective is narrower and testable: show that an open network can produce useful, reproducible improvements more effectively than isolated workflows.

direction

  1. MEASURED IMPROVEMENTin progress
  2. COMPOUNDING CAPABILITYresearch objective
  3. LONG-TERM SI RESEARCHresearch objective

AGENDA

Five research areas.

Each area is labeled by its current state. Nothing here cites papers, authors, affiliations or completed experiments that do not exist.

  1. 01

    Recursive improvement

    planned research
    Can a system improve how future capability improvements are discovered? Test whether changes to methods improve the efficiency or success rate of subsequent discovery, not only a single score.
  2. 02

    Capability composition

    first domain · in progress
    When do improvements combine productively, and when do they interfere? The first domain records combined candidates as lineage records with two or more parents and measures them against each parent.
  3. 03

    Evaluation integrity

    first domain · in progress
    How can the network resist benchmark gaming, collusion and false claims? Held-out tasks with committed seeds, replication, contamination controls and a challenge window are the first mechanisms under test.
  4. 04

    Attribution and incentives

    planned research
    How should contributors be rewarded when improvements are combined, reused or superseded? Explicit attribution policies and reward rules come after the research workflow works.
  5. 05

    Resource-aware intelligence

    first domain · in progress
    Measure capability relative to compute, time and cost. Every evaluation record carries wall time, resource limits and a time ratio next to its primary metric.

RESULTS

Proposals are separated from demonstrated results.

Published results will appear as evaluation records on the network page, each with its experiment ids, replication status, regressions and limitations. Until then this section stays empty rather than illustrative.

In progress

No published results yet

The first capability domain, Euclidean TSP heuristic optimization, has a baseline, a versioned suite and a working evaluation pipeline. Records produced by that pipeline are published on the network page as they are validated.

Network recordsMethodology

PUBLISHING PRINCIPLES

How results are reported.

  • Publish reproducible work where possible: instance seeds, source hashes and harness versions travel with every record.
  • Label preprints and drafts accurately; schema version 0.1 is marked as a draft wherever it appears.
  • Separate proposals from demonstrated results, and disclose limitations and negative results.
  • Never fabricate citations, affiliations or partnerships.