α · Research conduct charter

The research conduct charter.

Every contributor signs this on entry. Cells signing-in their contributors are responsible for ensuring the charter has been read and understood. The charter is not an indemnification document — it is a statement of structural commitments that we treat as the floor of the lab's operating culture.

Current version 4.2
Last revision 2024-02-11
Ratification path charter-class proposal (67% weighted threshold)
Prior versions archived in the internal index

Note: this is a public mirror of the operative charter. The signed-in version contains additional procedural appendices that are bound to specific tooling and only meaningful inside the lab. The public mirror contains all substantive commitments.

Article 1 · Standing and tenure

  1. I enter alphabell as a contributor in a cell that has accepted my onboarding. I am not an employee. The lab is not my employer of record.
  2. I will treat fellow contributors with the assumption of good faith, including disagreement. I will not use the protocol layer to harass, intimidate, or marginalize other contributors.
  3. I acknowledge that vote-of-record on lab-level proposals accrues at 24 months of signed contribution, and that this is a structural feature, not a hazing ritual.

Article 2 · Research practice

  1. I will publish to the internal index — at minimum, periodic updates that allow my cell's work to be reviewed by adjacent cells and by the long-tenured-contributor pool. The cadence is set by my cell; the obligation to publish is not.
  2. I will not present internal-index results externally without going through the publication policy. This includes informal external presentations, social-media commentary that contains substantive technical detail, and conversations with funders or partners about results that have not been released.
  3. If I observe a research result that I believe should be subject to the dual-use definition, I will raise it through the standard channel even if my cell disagrees. I will not be retaliated against for raising it.

Article 3 · Dual-use and the pairing requirement

  1. A research result is dual-use if it (a) materially increases the capability of a deployed AI system in a domain with foreseeable significant harms, including but not limited to: cyber-offensive applications, biological-design applications, large-scale social-engineering or persuasion applications, or autonomy in any safety-critical control loop; or (b) is so flagged by my cell's paired interpretability cell or by an axis steward, pending review.
  2. Any cell producing dual-use research will be paired with an interpretability cell with rolling read-access to its checkpoints and training logs. The pairing is not advisory. The interpretability cell may call a halt; the called halt will be honoured pending review.
  3. I will not work around an interpretability cell's halt by reconstituting the work outside the pairing scope. To do so is grounds for removal of standing.

Article 4 · Recursive self-improvement

  1. Any cell working on recursive-self-improvement-axis topics — including any cell whose work could be reconfigured into such a capability — operates under the modification-under-review (MUR) protocol described in axis publication 25/05 and its amendments.
  2. I will pre-register stopping conditions before any RSI run. Stopping conditions are non-negotiable during the run; modification mid-run is treated as a halt and a re-pre-registration.
  3. I acknowledge that the long-tenured-contributor quorum retains an indefinite hold provision for RSI-axis results that the paired interpretability cell flags. Such holds are reviewed annually.

Article 5 · Funding and conflicts

  1. I will declare any external funding I receive that could be construed as creating a conflict of interest with my cell's work. The declaration is to my cell, and to the long-tenured-contributor quorum if my cell asks.
  2. I will not accept a funder's specific instruction to direct alphabell's research toward an end the funder prefers. Funders are welcome to fund the lab; they are not welcome to direct it. Violations of this clause are charter-level violations.
  3. I will not represent alphabell as endorsing a funder's product, position, or organisation in any context where such endorsement would be construed as authoritative.

Article 6 · Compute and federation

  1. I will use compute that the scheduler has allocated to my cell, and not reallocate it informally. Trades of compute among contributors are not part of the protocol.
  2. If my cell receives unusual compute allocation — for example, by being routed through the RSI-axis isolated enclave — I will treat the special arrangement as binding and will not attempt to exfiltrate artefacts outside the enclave's audit scope.

Article 7 · Departure

  1. I may leave alphabell at any time, by signing-out through my cell. There is no exit interview obligation.
  2. I will not, after departure, present internal-index results or unreleased work externally for at least the duration of any active publication-delay window applicable to that work. The obligation is bounded; it is not a perpetual NDA.
  3. I retain the right to publicly describe my contributions to released work, and to discuss publicly the structural features of the lab. This is encouraged, not merely permitted.

Article 8 · Amendments

The charter is amended through the charter-class proposal mechanism described in /governance. The substantive provisions of this version have been amended three times since 2017. Procedural appendices are amended more frequently and do not require the charter-class proposal threshold.

Signature

By signing in to alphabell, I attest that I have read and understood the foregoing, and that I will treat its provisions as the floor — not the ceiling — of my participation in the lab.