RESEARCH · THE FRONT DOOR
Do some research
The programme is deliberately too broad for one person or one laboratory, and external replication is scientifically preferable to central ownership. The model is not outsourcing: MM defines the questions, the schemas and the integrity requirements; external teams execute bounded studies, and are entirely free to falsify the hypotheses. A researcher who convincingly kills a central claim has contributed to the programme.
UNDER CONSTRUCTION The full project listing is not published yet, on purpose. It goes live only after the delegation constitution is frozen: a short public document covering scope ownership, preregistration authority, scenario secrecy, authorship and contributor credit, data custody, safety review, publication rights, null-result rights, conflict disclosure, the sponsor firewall and correction obligations. Rules written after a collaborator has started are governance by negotiation; rules written before are governance by design. This gate is non-negotiable, and it protects you at least as much as it protects us.
What the listing will look like
Open questions grouped by delegation tier, each stating: the question · its status · what already exists · what we need · publication rights · a contact. Each carries a plain-language label: suitable for a Masters project, needs open-weight ML expertise, suitable for a psychometrics researcher, requires frontier-lab access, replication package available. Until it lands, the branch pages on the Research Tree already state every flagship study, kill criterion and delegation tier, which is most of the information a serious enquiry needs.
The four tiers
| TIER | SUITABLE WORK | LIKELY EXECUTOR |
|---|---|---|
| Tier 1 · protocolisable | Replications, paraphrase and translation robustness, scenario generation under schema, historical framing, Rung Zero corpus coding, public research-map updates. | Students, supervised research programmes, independent researchers. |
| Tier 2 · specialist | Human-agency studies, statistical validation, psychometrics, governance analysis, mechanism design. | University labs, Masters and PhD projects, domain specialists. |
| Tier 3 · technical collaboration | Open-weight interpretability, checkpoints, sparse-feature methods, J-space work, causal interventions, agentic sandboxes. | Mechanistic-interpretability and ML safety groups. |
| Tier 4 · gated frontier | High-capability control studies, successor work, persistence and replication analogues, advanced multi-agent asymmetry. | Frontier labs, government evaluators, secured partnerships. |
What never gets delegated
So expectations are set before anyone writes to us: the canonical definition of Keeping and the frozen claim boundaries; the core question and preregistration of flagship studies; control of active secret scenario pools and commit-reveal integrity; the canonical scoring schema and validation standards; interpretive authority over what a result does and does not establish; State of the Bridge release sign-off and corrections; and the sponsorship firewall. Everything else is on the table.
Independence, in writing
External researchers receive authorship according to substantive contribution; nobody is a contractor producing MM-branded results. Independent replications remain independent, including the right to publish a null or a contradiction. The programme becomes more credible, not less, when hostile or unaffiliated teams can run the instruments and disagree in public. Positive MM branding is never a condition of participation.
What you can do today, gate or no gate
Run the deposited battery: everything a rerun needs ships in the MMBP-1 deposit, and the Replication Club is open now. Tier 1 work will additionally ship with a packaged replication kit: frozen protocol, condition files where publishable, scoring schema, parser details, an example analysis command, the expected API-credit envelope and a one-page statement of what counts as an independent replication. Propose a scenario at the Dilemma Drop, attack the work at the red-team door, or open an early conversation at info@themeaningmotive.org with the branch you’re interested in; conversations can start before the constitution freezes, data collection cannot.
The ideal experiment is one worth losing. Come and lose one with us, or to us.