Research Disclosure & Reproducibility
The laboratory practices open science about its claims and protected engineering about its implementation. This page states where that line is drawn, so that readers can judge for themselves whether a given claim has been given enough support to be evaluated — and so that the laboratory cannot quietly move the line later.
Normally publishable, normally protected
These are defaults, not absolutes. A specific document may publish more than the first list or, for a stated reason, less.
Normally publishable
- research questions
- hypotheses
- experimental methodology at an appropriate level
- controls
- falsification criteria
- aggregate results
- limitations
- selected evidence
- experiment identifiers
- publication identifiers
- cryptographic evidence commitments where appropriate
Normally protected
- complete source code
- internal prompts and configuration
- credentials
- security architecture
- detailed infrastructure topology
- complete proprietary orchestration
- complete internal schemas
- private experimental biographies
- sensitive raw data
- information sufficient to compromise experimental integrity
- implementation details whose disclosure would unnecessarily permit reconstruction of proprietary laboratory infrastructure
Confidentiality is not evidence
A scientific claim cannot be made credible merely by asserting that supporting evidence is confidential. If the laboratory cannot publish enough for a claim to be evaluated, the honest options are to publish a weaker claim, to label the claim as unverifiable, or not to make it. Protecting an implementation is a reason to describe a result carefully; it is never a reason to ask for the reader's trust.
The same applies in the other direction: the laboratory does not treat the protected list as a legal category, and it makes no claim here about what is or is not a trade secret. The list describes what the laboratory will and will not ordinarily publish, nothing more.
Verification on request
Selected underlying evidence may be made available for legitimate verification when appropriate — for example to a reviewer, a collaborator, or a researcher attempting to check a published claim — subject to safety, privacy, experimental-integrity, legal, and proprietary constraints. Requests can be sent to research@growingresearch.org. The laboratory will say plainly when it declines, and why, at the level of the categories above.
Reproducibility without the platform
Because the platform itself is not released, an outside party cannot currently reproduce a Second World experiment end to end. The laboratory's approach is to make each published claim independently testable in other ways where practical: by publishing the methodology and controls at the level needed to evaluate the design, by committing to retained evidence with cryptographic hashes at the time of publication, by using stable public experiment identifiers so that later documents refer to the same history, and by stating explicitly which claims rest on evidence that has not been released.