# CrystalRAM Research Commons governance charter

Status: operational charter for the private research phase. Public release,
formal incorporation, trademark assignment, and licensing require separate
explicit decisions by the project steward and contributors.

## Purpose

CrystalRAM exists to discover, test, and document useful memory architectures
for AI. Its scientific record belongs to the collaboration that produces and
reproduces it. Participation by a company, university, laboratory, model
provider, or funding source does not convert the project into that entity's
exclusive accomplishment.

## Independence principles

1. **Claims belong to evidence.** A scientific claim is attributed to its
   recorded contributors, sources, methods, and reviewers—not to the vendor of
   the AI model used to assist the work.
2. **Providers are provenance, not authority.** Model and provider names are
   recorded so work can be reproduced and biases examined. They grant no
   scientific rank, ownership, or veto.
3. **No exclusive credit.** Project communications must not imply that one
   company, model, institution, or sector created CrystalRAM when the record
   shows a multi-party contribution.
4. **No pay-to-win science.** Funding may support experiments but cannot alter
   evidence status, suppress negative results, or bypass review gates.
5. **Open participation criteria.** Participation is based on compliance with
   the scientific and safety protocols, not commercial membership.
6. **Conflict is preserved.** Contradictory results remain visible until a
   distinguishing experiment resolves them.
7. **Human accountability remains.** AI systems may investigate, propose, run,
   and review work, but consequential project governance and external
   publication remain subject to human approval.

## Attribution

Every finding and review records the stable contributor identity, contributor
kind, role, organization when disclosed, AI provider and model when applicable,
code revision, evidence, and artifacts. Provider/model attribution describes
the instrument used; it does not make the provider an author unless that entity
separately contributed people, funding, data, equipment, or intellectual work
and that contribution is explicitly recorded.

Project-level summaries must credit the human and AI collaboration and link to
the finding-level record. A contributor may request pseudonymous attribution,
but cannot use anonymity to conceal a material financial or institutional
conflict relevant to a claim.

## Cross-boundary review

For findings proposed as `supported`, the project should seek at least one
review across an independence boundary whenever practical:

- a different AI provider or independently operated local model;
- a different company, university, or laboratory;
- or a qualified human reviewer not involved in the original work.

Cross-provider review is not automatically independent if the same operator,
prompt, code defect, dataset, or hidden assumption is shared. Review records
must identify what was genuinely independent.

## Decision boundaries

- Scientific status follows `research/PROTOCOL.md` and cannot be changed by a
  branding, funding, or governance vote.
- Routine code changes follow pull-request review and automated verification.
- Changes to evidence rules, attribution rules, governance, licensing,
  repository visibility, or external publication require human project-steward
  approval and a recorded rationale.
- No automated agent may grant credentials, change access control, publish the
  repository, or merge its own scientific finding without human authorization.

## Anti-capture safeguards

- No single organization should hold more than one third of seats on a future
  scientific steering council.
- Council membership should span at least three sectors, such as device
  physics, photonics/materials, computer architecture, AI systems, academia,
  public-interest research, and independent engineering.
- A member with a material conflict must disclose it and abstain from the
  affected governance decision.
- Scientific records and event history must remain exportable in documented,
  non-proprietary formats.
- The canonical schemas and API must remain implementable without a proprietary
  AI SDK or hosted service.

## Amendments

Governance amendments use a pull request that states the problem, proposed
change, affected stakeholders, and anti-capture analysis. During the private
phase, the human project steward approves amendments. Before public expansion,
this charter should be reviewed by contributors and qualified open-source and
research-governance counsel.