This directory holds the community-growth strategy, its measurement models, and
the ACE (Autocatalytic Community Evolution) design stack, plus the bounded
evidence records produced against them.
Read the authority line first. Every ACE layer below declares itself
experimental, shadow-mode, proposed, or offline, and
the Phase-B activation gate below states “Authority: none by
itself” and that a controller cannot activate itself. Nothing in this directory
grants an autonomous actuator. The presence of a complete design stack is not
evidence that the stack is switched on.
The layers are separate on purpose: causal evidence, denominator inventory, and
value/cost measurement must not collapse into one object. The ordering below is
the one the documents themselves describe.
Workflow-owned tracking issues
Two workflows persist state in GitHub issues, but they do not currently recover
from closure in the same way:
ace-community-growth.yml finds the ledger by label ace:ledger among
state: 'open' issues (issue 23). Closing it can still make the workflow fall
back to seed state and create a duplicate ledger; reopening the original after
that can trigger the explicit multiple-open-ledger guard.
ace-cohort-observer.yml identifies the observatory by label
ace:cohort-observer across all issue states (issue 109). A closed canonical
observatory is reopened and reused instead of replaced, so closure no longer
forks the observation series into a new issue number.
Both issues remain workflow-owned state containers whose bodies are rewritten on
each run. They are not ordinary tasks and should not be closed as part of an
issue-tidying pass. The observer is now resilient to accidental closure; the
community-growth ledger still depends on staying open until its own lookup is
hardened.
Strategy and measurement models
Contribute with your own AI coding agent — how to point Claude Code, OpenAI Codex, or Google Jules at AGENTS.md and reach a reviewable pull request, with paste-ready briefs, the checks to run, and the provenance note every agent-assisted change carries.
Contributor Quickstart — the shortest path from visitor to a bounded first contribution, with live newcomer issues across documentation, testing, reproducibility, and external-machine validation.
Contributor Experience Guidelines — practical norms for turning curiosity into autonomous, repeat participation without dark patterns, spam, fake urgency, or deceptive social proof.
Community Growth Strategy — why community
growth is a parallel product/systems problem rather than a post-engineering
marketing phase, and the contribution forms it needs.
Visibility + Community Self-Growth Loop — makes discoverability, SEO/AEO, GitHub interest, contributor acquisition, retention, and reviewer/leader growth explicit self-improvement axes while keeping popularity separate from correctness authority.
Community Growth Dynamics — research model and
measurement specification; turns growth into a dynamical-system problem over an
explicit community state, without treating stars or raw contributor count as fitness.
ACE: GitHub-Constrained Self-Improving Community
— working design for the next ACE iteration under real GitHub constraints
(issues, PRs, labels, Actions, API limits), explicitly not publicity
automation.
ACE Activity Metabolism — the composition rule
that turns repository activity into verified capability, knowledge, repair, or
reproductive opportunity. Activity itself is not fitness.
ACE evidence and control layers
Ordered from raw observation to the gate that could, in principle, leave shadow
mode. Each is narrow by design and cites the issues it was built against.
ACE Cohort Observer v0 — experimental,
metadata-only observability. Separates activity, cohort exposure, candidate
work, and causal evidence. Built against #40.
ACE Lineage Protocol v0.1 — the smallest
GitHub-native record for a verified parent -> Growth Seed -> candidate
descendant -> verification -> verified descendant chain. Lineage evidence
only, not a database or an actuator. Related: #10, #23, #25, #27.
ACE Live Carrying-Capacity Model — proposed bootstrap
model; makes the ecological capacity gate depend on current recoverable
review pressure instead of a cumulative event counter.
ACE Generation Evidence Interface —
Phase-A / shadow-controller interface holding parents, lineage receipts, and
measurement as three independent layers. Related: #25, #40, #57.
ACE Phase-B Activation Gate — experimental,
fail-closed, offline conjunctive rule for whether the repository is allowed and
evidenced well enough to leave shadow mode. No authority by itself.
Bounded experiments and evidence records
Dated records measured against the designs above. Each is a snapshot at its
stated window or revision, not current status.
ACE Bootstrap Experiment — Cohort 1 — active
bootstrap experiment over the 2026-08-28 to 2026-09-27 window, five Growth
Seeds; public state tracked in issue #23.
Newcomer Path Audit —
2026-08-29 first-contact walk from the public README at 63e4acc to a
realistic bounded task, with approximate navigation times.
Starter Tasks — ten bounded contribution opportunities
across testing, security, documentation, research, review, community and
tooling, each checked against the tree at 2ce141a and carrying its own
acceptance test. Written because every newcomer-labelled issue at that
revision was an independent-review request.
Free Development Automation Resources
— zero/low-cost agent tooling comparable to the repository’s Jules
automation (OpenHands, Gemini CLI, mini-SWE-agent, goose, and others),
verified against vendor documentation on 2026-09-22. Agent output remains a
candidate change, never acceptance evidence or merge authority.
Adding a document
State the document’s status and authority in its own header, place evidence
records under onboarding-tests/ or task-decompositions/ rather than at the
directory root, and add one line to the group above that it belongs to. A new ACE
layer must say what it does not authorize.