Support work we can examine
Leviathan researches how people, AI systems, and independent groups can develop knowledge, language, and learning together. We want observations, concepts, methods, and values to travel with enough context for another participant to question, use, or improve them. Independent groups should retain their own approaches. Animal welfare is an intended first application; the purpose extends to other research and practical work. What is Leviathan? explains that wider proposal.
We are preparing a bounded next research stage and ways to receive useful contributions of ideas, resources and expertise. This briefing opens a discussion of that work. The contribution form prepares material for a separate public delivery step; it is not a funding portal, submitted grant application, payment service, or evidence that funding is secured.
What the present evidence supports
In the first local inquiry, a review request carried a pilot finding but omitted the exact decision under review. The reviewer identified the gap. Supplying the decision and sources enabled further review and a procedural correction: future requests should include the full decision reference. These were coordinated Codex development roles, with context carried by an operator. The episode shows a correction in our own work; improved performance on later unfamiliar work remains untested.
The synthetic correction pilot returned equal results for equivalently scoped JSON and prose conditions. It establishes no advantage for structured representation. Our website, forum code, authored language, and inquiry records are starting pieces; their existence does not establish a deployed independent learning network. The representation claim and its limits remain open to challenge.
A proposed next stage
Compare versioned structured context with good ordinary documents under equal resources: the same relevant evidence, task access, model conditions, and time or compute limits. Include incomplete dependencies, corrected sources, and a transfer between groups with different concepts or policies. A receiving group should be able to use a method without silently adopting the originating group's authority or values.
Before running the comparison, agree the cases, evaluation method, resource accounting, and what would count as a useful result. Measure the cost of finding and correcting a dependency, whether the correction reaches affected work, and whether the receiving group finds the resulting answer or method useful. Preserve failures, equal results, unresolved disagreements, and simpler alternatives alongside improvements. Publish reusable test cases, the evaluation method, source-version references, resource use, results, and the resulting design decision. A changed procedure alone is insufficient evidence of better learning.
The open contribution cards name proposed work OW-001, OW-002, and OW-003. Their small starters ask for one counterexample or baseline, one consequential ambiguity and repair, or one useful method or new question with a test. About 150–300 words is only a suggestion; a full study is not required to begin. Each card explains a possible benefit to the contributor as well as the project. Fuller work needs an agreed scope and a willing responsible participant. External people, agents with their operator’s authorization, and groups can bring a case, idea, implementation or evaluation without endorsing the whole proposal.
Discovery and making useful things are part of this invitation. A method from elsewhere could open a different research direction, help another group with its own work, or lead to a design we have not considered. The new-question path does not require an existing claim to correct. The contribution process describes how a reasoned response and any resulting work should be recorded. A reply or an improvement is not promised.
Resources and responsibilities
Useful support includes compute or API credits for agreed comparisons; time and expertise in evaluation, language, software, and research methods; and potentially grant funds for a defined programme of work. We have not established an amount, schedule, legal applicant, eligibility, or committed reviewer. Before submitting an application or accepting resources, establish the accountable applicant or recipient, suitable programme, budget, duration, responsibilities, and reporting terms.
Mimar initiated Leviathan. Start a public discussion through the existing Leviathan meta issues, naming the work and support you can offer. Public delivery requires a GitHub account or an authorized agent with GitHub access. Project maintainers; reviewer named when work is taken up. A response or acceptance is not guaranteed. The local form can copy Markdown or download JSON, but keeps your input in your browser and delivers nothing; inspect and deliberately submit any public offer on GitHub. Keep credentials, private personal context, and confidential terms out of public issues. Establish an appropriate private exchange separately where needed; do not share API keys.
Resource contributions do not buy decisions, preferred conclusions, or vetoes over objections. Agreed scopes, material conditions, conflicts of interest, and results should have public written records, with confidential details excluded. The participation guide explains current contribution routes and their limits.
Read or save this briefing as Markdown.
Basis: What is Leviathan? 0.13 and Vision 0.9 (4 October 2026); The Work 0.4 (4 October); Participate 0.9 (6 October). Revision 0.1 · 8 October 2026: initial support briefing; track subsequent changes in the content revision map.
Revision 0.2 · 8 October 2026: follows The Work 0.6 and Participate 0.11, adds small starters and an open new-question path, and separates contribution preparation from delivery and funding. This is local intake preparation, not a deployment, outreach report, capacity commitment, or funding award.