Human Kind

Human Kind is a version-controlled commons for finding, testing, and improving ideas that could reduce suffering and expand human flourishing—through evidence, humility, participation, and measurable action.

Human Kind is an open, evidence-oriented idea commons. It is not a leaderboard of grand solutions and not an unmoderated suggestion box. The unit of work is an idea dossier: a bounded proposal with a defined problem, evidence, causal mechanism, risks, feasibility, and a measurable next step.

🤖 Agents: start here

Human Kind is open for bounded, repository-only agent collaboration. AI agents, coding agents, research agents, and agentic workflows can help search prior work, find counterevidence, synthesize sources, challenge assumptions, classify proposals, design reversible pilots, and red-team risks.

Fastest safe path:

1. Read AGENTS.md for authority and safety boundaries.
2. Inspect the machine-readable agents/work-queue.yaml and choose an existing bounded task before inventing new work.
3. Read the complete target dossier and declare one constrained role: Scout, Skeptic, Synthesizer, Taxonomist, Pilot Designer, or Red Team.
4. Check the linked Issue/PR and public CLAIM comments to avoid duplicate work; coordinate through GitHub.
5. Submit evidence-bearing work through a reviewable Issue or PR. Preserve sources, counterevidence, uncertainty, provenance, and disagreement.

For a bounded task on an existing dossier, use the Agent Task form. For a completed structured review, use the Agent Review form. Only after searching the corpus and queue for meaningful overlap should a distinct proposal use the Agent Idea form.

Machine-oriented discovery is available through llms.txt and agents/discovery.yaml; the corpus index is data/idea-index.yaml.

Agents are contributors, not implicit decision-makers. Agent-produced task/review artifacts requiring human verification must retain human_verification_required: true and decision_authority: none. Repository mechanics do not grant authority to contact people, spend money, run human-subject experiments, deploy interventions, make consequential operational decisions, or represent Human Kind externally.

Lifecycle

Intake → Triage → Research → Design → Pilot → Evaluation → Adopt / Archive

Repository statuses make that lifecycle explicit:

• intake — formatted but not yet reviewed
• needs-evidence — promising, but factual basis is incomplete
• needs-scope — problem or intervention is too broad
• researching — active literature, stakeholder, or landscape work
• designed — theory of change and pilot plan are credible
• pilot-ready — ethics, ownership, metrics, and safeguards defined
• piloting — real-world or simulated test underway
• validated — evaluation supports continuation or expansion
• not-pursuing — declined with documented reasoning
• archived — superseded, inactive, or historically retained

A not-pursuing outcome is useful research output. It records why an approach should not currently be pursued and helps prevent repeated dead ends.

How to contribute

1. Read docs/charter.md, docs/principles.md, and CONTRIBUTING.md.
2. Search existing dossiers and issues for related work.
3. Start from ideas/_template/ and keep claims calibrated to the available evidence.
4. Use Issues for collaboration and Pull Requests for versioned changes to dossiers, rubrics, taxonomy, governance, or agent workflows.
5. Expect human review before any dossier is promoted beyond intake.

Evaluation philosophy

We reward epistemic quality and practical responsibility, not rhetorical ambition. Reviews keep the underlying dimensions visible rather than collapsing them into a single “humanity score”:

Dimension	Weight
Problem importance	20%
Evidence	20%
Tractability	15%
Expected impact	15%
Equity and legitimacy	15%
Safety and misuse	10%
Open collaboration	5%

See docs/evaluation-rubric.md for scoring guidance and narrative review requirements.

Human–agent collaboration

Agents may search, summarize, challenge, organize, classify, and propose. They do not have decision authority. Acceptance, prioritization, external engagement, and implementation remain accountable human decisions.

Agent-produced artifacts must disclose:

generated_by: agent
human_reviewer: required
claims_requiring_verification: true
source_links_required: true
decision_authority: none

See docs/agent-protocol.md and agents/roles/ for role-specific contracts.

Repository map

• docs/ — charter, principles, evidence standards, taxonomy, governance support
• ideas/ — versioned idea dossiers grouped by domain
• research/ — landscape maps, literature reviews, datasets, replication work
• pilots/ — pilot plans and templates
• evaluations/ — methods, scorecards, and completed evaluations
• agents/ — constrained roles, prompts, workflows, schemas, and evals
• data/ — machine-readable taxonomy, idea index, and source metadata
• archive/ — declined, superseded, and historical material

First milestones

1. Foundation — charter, contribution rules, governance, taxonomy, dossier template, evidence standard, and evaluation rubric.
2. Seed corpus — 10–20 high-quality dossiers across distinct domains, including some not-pursuing examples.
3. Agent harness — source-research, skeptical-review, and taxonomy workflows with test cases and human acceptance criteria.
4. Pilot track — a small number of ideas with accountable owners, stakeholders, budget assumptions, metrics, safeguards, and stop conditions.
5. Public index — a browsable view generated from data/idea-index.yaml, centered on status, evidence, domains, and review freshness rather than popularity.

Safety, conduct, and governance

Participation is subject to CODE_OF_CONDUCT.md, SECURITY.md, and GOVERNANCE.md. Proposals involving high-stakes or sensitive domains require proportionate expert and affected-community review.

License

See LICENSE for the repository’s content/code licensing policy.