selfdriven · Science

We don't prove theories. We make them testable.

An open, distributed home for adversarial, reproducible research. Hypotheses become preregistered experiments. Experiments become independent implementations. Evidence becomes a graph — including the parts that fail.

Explore CBR-OEN How the network works
EVIDENCE GRAPH / LIVE VIEW A claim, its experiments, and the challenges it has survived.
10

preregistered experiments in CBR-OEN, E001–E010

7

participant roles, each performable independently

R3

replication target — three independent institutions

v0.1

open protocol, shared by every network

The founding commitment

Where a claim can become an experiment, test it openly.

Where a claim about how the world works can be turned into a reproducible experiment, it should be tested openly, independently and adversarially. No organisation owns the conclusions.

No experiment proves the whole theory.

Read the principles
Specify PREREGISTERED · FROZEN sha256:9f… Spec Frozen Hashed

Testable before argued

Hypothesis, null, rules, measurements and analysis are written down and hashed before any results exist.

  • Preregistration
  • Falsifiable
  • Provenance
Implement INDEPENDENT CODE · ONE SPEC Spec Implementations Reproduced

Independent by design

Confidence grows when independent programs — not shared code — reproduce the same phenomenon, in any language.

  • Implementation
  • Replication
  • Any language
Challenge STRONGER BY SURVIVING Challenge Claim Stronger

Adversarial on purpose

Review hunts for artefacts and hidden assumptions. A good challenge is a contribution — and a failed replication is kept, not dropped.

  • Adversarial review
  • Negative results
  • Open

The lifecycle of a claim

From a sentence to a replicated finding.

Every result travels the same path. Each stage can be performed by a different group, at a different institution — and each can fail independently, which is itself a result worth keeping.

  1. 01 · Specify

    Preregister

    Hypothesis, null, rules, measurements and analysis are written down and hashed before results exist.

  2. 02 · Build

    Implement

    Independent teams build to the specification — not by copying the reference code — in any language.

  3. 03 · Run

    Execute

    Registered runs execute across university, HPC and independent nodes. Each emits a signed manifest.

  4. 04 · Analyse

    Replicate

    Blinded analysis and independent replication test whether the effect survives new hands and new code.

  5. 05 · Contest

    Challenge

    Adversarial review hunts for artefacts and hidden assumptions. Surviving a challenge strengthens the claim.

The network

Projects, principles, and how the federation runs.

Each project is a federation, not a lab — a shared protocol that many independent groups can execute, replicate and contest.

Projects

Collaborations on the network.

The first network is open for participation. The protocol is domain-agnostic — more will follow.

Principles

Ten principles. One that matters most.

The commitments every participant shares — whatever they believe about the theory being tested.

  • 01

    Testability before advocacy

    Claims are expressed as falsifiable propositions wherever possible — not positions to be defended.

  • 02

    Specification before execution

    The experiment is the spec. It is defined and frozen before any results are observed.

  • 03

    Independence of implementation

    Confidence grows when independent programs — not shared code — reproduce the same phenomenon.

  • 04

    Separation of roles

    Hypothesis, implementation, execution and analysis are kept apart wherever practical.

  • 05

    Negative results are first-class

    A finding that fails to reproduce is retained, indexed and discoverable — never quietly dropped.

  • 06

    Provenance is evidence

    Software version, parameters, seeds and analysis procedure are integral parts of a result.

  • 07

    Criticism is encouraged

    A result should become stronger by surviving challenge. A good challenge is a contribution.

  • 08

    Institutional autonomy

    Participating institutions stay scientifically independent. Contribute one element or many.

  • 09

    Open evidence by default

    Specifications, implementations, results and critiques default to public availability.

  • 10

    No experiment proves the whole theory

    Evidence applies to individual claims, not to a worldview.

Governance

Eight areas of focus, adapted for open science.

The selfdriven Foundation organises every entity around the same eight areas — coordinated by human conductors, executed by many independent hands.

  • 01

    Direction

    Research program, founding questions and the protocol roadmap.

  • 02

    Engagement

    Institutional partnerships, adversarial collaboration and invitations.

  • 03

    Enablement

    Specifications, test vectors and onboarding for independent implementers.

  • 04

    Protocols

    Experiment schemas, run manifests, provenance formats and the evidence graph.

  • 05

    Sustainability

    Open infrastructure, compute federation and long-term archival.

  • 06

    Processes

    Preregistration, execution, replication and the statistical protocol.

  • 07

    Accountability

    Reproducibility status, negative-result retention and challenge records.

  • 08

    Organisational

    Federation of working groups, the Protocol Council and decentralised nodes.

The evidence graph

Papers are views. The graph is the record.

selfdriven.science doesn't treat the paper as the fundamental unit of knowledge. Results live in a graph of typed relationships — so a disagreement stays represented instead of collapsing into a single consensus sentence.

  • PREREGISTERED
  • EXECUTED
  • INDEPENDENTLY_REPRODUCED
  • INSTITUTIONALLY_REPLICATED
  • MATHEMATICALLY_REVIEWED
  • CHALLENGED
  • CONTRADICTED
  • SUPERSEDED
  • Run RS-1098SUPPORTSCBR-H003
  • JULIA-01REPLICATESPY-REF-03
  • RUST-04FAILS_TO_REPLICATECBR-E005
  • CBR-C017CHALLENGESCBR-E003
  • MR-19REVIEWSCBR-E004
  • CBR-E003:v1.2SUPERSEDESv1.1

selfdriven.science · Participate

Contribute one element. Keep your independence.