Skip to content

COLLECTIVE INTELLIGENCE EVOLUTION

Intelligence should compound.

An open protocol for discovering, validating, and inheriting improvements in intelligence, coordinating independent systems toward increasingly capable collective intelligence.

Explore the ProtocolRead the Research

  • Network

    Robinhood Chain · 4663

  • Evaluation

    Independent, reproducible

  • Records

    Lineage-preserving

  • Status

    Mainnet deployed

GET /api/v1/evaluations/eval_2f91
200 OKIllustrative
{
  "id": "eval_2f91",
  "proposal": "prop_7a1c",
  "parents": [
    "bl_0412"
  ],
  "baseline": "bl_0412",
  "suite": "suite_tsp-v3-1",
  "status": "VALIDATED",
  "primary_metric": {
    "key": "mean_tour_length",
    "baseline": 1,
    "candidate": 0.9412,
    "relative_delta": -0.0588
  },
  "reproduction": "3/3",
  "summary": "Shorter tours on dev and held-out sets. 1.4x baseline time. No invalid solutions."
}
improvement recordVALIDATED

CONCEPTUAL MODEL OF INTELLIGENCE EVOLUTION

EVOLUTION CYCLE

One improvement, start to finish.

Six stages carry a candidate from proposal to inheritance. Each stage leaves a record that later work can trace, reproduce, and build on.

traceexp_3c7d
completeIllustrative

stage 01 · Propose

Submit a candidate improvement to a defined capability, algorithm, reasoning method, or research procedure.

proposal: prop_7a1c · parents: bl_0412

CENTRAL THESIS

From isolated capability to collective evolution.

Today, intelligence systems are often developed and improved in separate environments. Useful discoveries can remain isolated, while reproducing results, attributing contributions, and building upon previous improvements require separate coordination.

The protocol proposes a shared process for that work. Independent participants propose changes. Reproducible experiments test them. Independent evaluators assess the results. Validated improvements become reusable components in a growing network of intelligence research.

The objective is not merely to produce better answers, but to improve how future capabilities are discovered.

Isolated discoveriesShared, reproducible lineage

INTELLIGENCE LINEAGE

Every improvement has a lineage.

Improvements should not be disconnected benchmark scores. Each accepted contribution preserves its relationship to earlier work, the conditions under which it succeeded, and the components that made the final result possible.

lineagebl_0412→ 6 records
Illustrative data
bl_0412BASELINEprop_7a1cVALIDATEDprop_7a2eREJECTEDprop_7a3fVALIDATEDprop_8b10VALIDATEDprop_8c22PENDING
baseline validated pending · in progress× rejected · superseded (dimmed, never erased) combines

DIFFERENTIATING MECHANISM

Improvement is not a claim. It is a result to reproduce.

A proposed change should not become a recognized improvement because its creator reports a better score. The protocol is designed around controlled comparisons, independent evaluation, held-out tasks, reproducibility, and explicit records of how results were obtained.

  1. 01

    Measured improvement

    baseline vs candidate

    Compare a candidate with a defined baseline using preregistered criteria. A self-reported score is not a result.

  2. 02

    Generalization

    held-out tasks

    Test whether gains persist on held-out or unfamiliar tasks whose instances were committed before the run.

  3. 03

    Efficiency

    compute · time · cost

    Track performance alongside compute, execution time and cost, so a gain bought with ten times the budget is reported as such.

  4. 04

    Recursive value

    design goal · not yet measured

    Evaluate whether a change improves the ability to discover further useful improvements, not only a one-time score.

Explore the evaluation methodology →Design goals
comparisonprop_8c22
Design goalsIllustrative

baseline

bl_0412

candidate

prop_8c22

Mean tour length (normalized)

1.000

baseline

0.927

candidate

−7.3%

lower is better

Generalization
dev −7.9% · held-out −7.3% · gap +0.6 pt
Compute
1 vCPU · 128 MB
Time
0.82 s → 1.31 s (1.6×)
Test suite
suite_tsp-v3-1
Reproduction
2/3 replications agree
Known regressions
  • p95 time +22% on n=1000 instances
statusPending replication

NETWORK STATUS

Coordination you can verify.

Chain-verified data and off-chain service reports are kept apart. If a service is not live, the panel says so instead of showing simulated numbers.

Open the network page
network statusRobinhood Chain · 4663
Mainnet deployedindexer · Synced

  • Proposals

    7

  • Completed experiments

    13

  • Pending replications

    0

  • Validated improvements

    6

chain integrity

Latest indexed block
91,107,071
Chain head
91,107,076
Indexer sync
Synced
Last synchronization
Oct 17, 2026, 05:35 PM UTC
Deployment version
v0.1.0
Registry
0xDC95f1…59de64 · verified

research activity

Supported domains
1
Challenged results
0
Evaluation service
Sandbox configured
Job queue
inline
Records database
Reachable
Generated
Oct 17, 2026, 05:35 PM UTC

Chain-verified values come from indexed contract events; service values are off-chain reports.

ARCHITECTURE

Intelligence evolves off-chain. Coordination is verifiable on-chain.

Intelligence research needs flexible execution environments and substantial computation. The protocol separates off-chain experimentation from on-chain economic coordination. Robinhood Chain supports settlement, commitments, and contribution accounting; it does not execute training or decide whether an improvement is valid.

  1. 01

    Intelligence layer

    Candidate algorithms, model changes, research procedures, and experiments.

    off-chain · flexible execution
  2. 02

    Evaluation layer

    Baselines, held-out tasks, independent replication, and result reporting.

    off-chain · sandboxed · reproducible
  3. 03

    Protocol layer

    Improvement records, lineage, attribution, challenges, and incentive rules.

    off-chain registry · on-chain commitments
  4. 04

    Settlement layer

    Production smart contracts on Robinhood Chain · 4663: commitments, challenge status and, later, contribution accounting.

    on-chain · verifiable
Intelligence → settlementHover a layer

Large artifacts, datasets, model weights, logs and anything sensitive stay off-chain. On-chain state is limited to record ids, content commitments, statuses, contributor addresses and settlement.

RECORDS

Every result has a shape.

Proposals, evaluation records and lineage records are structured documents. Their canonical hash is what the on-chain registry commits to, so anyone can check that a published record matches its commitment.

Draft schema · subject to change
{
  "schemaVersion": "0.1",
  "id": "prop_7a1c",
  "capabilityId": "cap_euclidean-tsp-v1",
  "baselineId": "bl_0412",
  "parents": [
    "bl_0412"
  ],
  "contributor": "0x3f0c9e21a7b4d5e6f70819a2b3c4d5e6f7089a1c",
  "title": "Or-opt refinement after construction",
  "summary": "Applies an Or-opt local search pass after nearest-neighbor construction. Deterministic given the harness-provided seed.",
  "method": "Relocate segments of length 1 to 3 whenever the move shortens the tour; iterate to a local optimum.",
  "language": "javascript",
  "entry": "solve",
  "sourceHash": "0x6f1d2c3b4a59687706f5e4d3c2b1a0998877665544332211ffeeddccbbaa9988",
  "createdAt": "2026-10-17T09:30:00.000Z"
}
  1. 01

    Define a baseline

    Pick the capability, the reference implementation and the versioned test suite the candidate will be measured against. The baseline is fixed before anything runs.

  2. 02

    Run a reproducible experiment

    The candidate and the baseline execute in an isolated sandbox with the same instances, seeds and limits. Inputs, versions and resource usage are captured with the output.

  3. 03

    Submit for independent evaluation

    Replication runs reproduce the result before the record is validated. Rejections, regressions and limitations become part of the permanent record.

OPEN PARTICIPATION

No single intelligence system should define the network.

Participants should be able to develop competing approaches, test alternatives, reproduce results, and build upon validated work without the network adopting a single model provider or intelligence lineage.

01

Researchers

Propose algorithms and capability improvements against defined baselines.

02

Builders

Implement and combine validated improvements into new candidates.

03

Evaluators

Reproduce results, run replications and identify regressions.

04

Sponsors

Fund defined research objectives once funding rules exist.

05

Infrastructure contributors

Provide suitable, isolated experiment environments and evaluation capacity.

These are participation roles, not a promise that every role will be operational at launch.

ROADMAP

Five stages, truthful status.

A user interface alone never completes a milestone. Statuses come from configuration and the verified deployment manifest, so the roadmap cannot claim more than the system does.

Full roadmap
  1. 01IN PROGRESS

    Research foundation

    Choose the first capability domain, establish a baseline and evaluation methodology, define improvement and lineage schemas, build the initial experiment runner.

    schemas v0.1 drafted · runner implemented · methodology under review

  2. 02IN PROGRESS

    Reproducible improvement

    Accept proposals, run controlled comparisons, support independent reproduction, publish evaluation records and their limitations.

    pipeline implemented · no public results yet

  3. 03PLANNED

    Open contribution network

    Add attribution, independent evaluator participation, research funding and reward rules, and a formal challenge process.

    design stage

  4. 04MAINNET DEPLOYED

    Mainnet coordination

    Deploy the approved minimal contract set to Robinhood Chain Mainnet, verify source and addresses, integrate settlement, test production flows, publish security and operational status.

    KuxerRegistry 0xDC95f1569E0399114D6764dF5a1A10603759de64 verified · release v0.1.0

  5. 05PLANNED

    Recursive evolution

    Test combinations of validated improvements, evaluate improvements to discovery methods, measure repeatable benefits, expand research domains only when evidence supports it.

    research objective

PLANNED · IN PROGRESS · STAGING · MAINNET DEPLOYED (contracts deployed and verified) · LIVE (operational end to end)

FAQ

The details.

Short answers to the questions that decide whether a protocol like this is credible.

BUILD

Help build the infrastructure for intelligence that evolves.

Contribute research, build evaluation tools, reproduce results, or help develop the protocol's technical foundations.

Explore DocumentationContribute to Research
Illustrative response
$ curl -s https://kuxer.tech/api/v1/lineage/lin_9d4e
{
  "id": "lin_9d4e",
  "child": "prop_8b10",
  "parents": [
    "prop_7a1c",
    "prop_7a3f"
  ],
  "relation": "combines",
  "evaluation": "eval_6e07",
  "commitment": "0x8f3a…c1e2",
  "status": "VALIDATED"
}