IDKMesh

ACE Cohort Observer v0

Status: Experimental, metadata-only observability.

This observer exists because IDKMesh must distinguish activity, cohort exposure, candidate work, and causal verified-descendant evidence.

Its role in the ACE stack is intentionally narrow:

ACE Cohort Observer (#40)
  -> trusted Bootstrap Cohort inventory
  -> external-interest / claim / candidate funnel
  -> bootstrap exposure denominator + review-capacity context

ACE Lineage Protocol (#48 / #25)
  -> parent -> seed -> descendant causal receipt
  -> verification evidence

ACE generational controller (#68 / #57)
  -> consumes validated evidence
  -> learns bounded community strategy in shadow mode

The observer must not duplicate the lineage parser or claim the full community reproduction number from labels alone.

State funnel

For each trusted issue labelled growth-seed in Bootstrap Cohort 1, the observer tracks:

trusted visible seed
  -> external interest
  -> claim
  -> candidate PR
  -> merged candidate
  -> bootstrap verified-descendant signal

Cohort admission requires all of:

This prevents untrusted marker text from manufacturing cohort membership even before the broader ACE workflow-hardening PR is integrated.

External interest

A non-owner, non-bot contributor comment or contribution signal associated with the seed. Interest is diagnostic evidence, not a descendant.

Claim

A seed is considered claimed when an external non-bot contributor is assigned, comments /claim, or authors a PR that GitHub cross-references from the seed. /claim grants no repository permission or exclusive ownership.

Candidate PR

An external-human-authored PR cross-referenced from the Growth Seed. A candidate is not automatically useful or verified.

Bootstrap verified-descendant signal

A candidate PR enters the observer’s bootstrap verified set only when it is merged and carries ace:verified-descendant.

This label is an explicit trusted acceptance signal. It is not equivalent to the full ACE lineage record. After the lineage protocol is accepted, a real causal claim should be represented by a validated ACE_LINEAGE receipt with verification evidence. The label can support that receipt; it cannot replace it.

Metric scope

The observer reports:

SeedReproductionRatio
  = bootstrap verified descendant PRs
    / trusted Bootstrap Cohort Growth Seeds

This is a bootstrap exposure metric, not the full R_community.

The full quantity remains conceptually:

R_community(W, t)
  = verified causal descendants
    / eligible matured verified parents

Computing it requires two independent inputs:

  1. a denominator inventory that keeps zero-descendant eligible parents visible and handles right-censoring;
  2. validated causal lineage receipts from the ACE lineage contract.

The cohort observer contributes useful denominator/exposure evidence for the bootstrap experiment, but its Growth Seeds are not automatically identical to all future verified parents. It therefore sets full_r_community_ready: false in its snapshot and must not overclaim causal reproduction.

Capacity-aware expansion recommendation

The observer may report HOLD_COHORT_1 or EVALUATE_COHORT_2.

EVALUATE_COHORT_2 currently requires at least two bootstrap verified descendant PRs, at least two claimed seeds, no more than three open external candidate PRs, and ACE capacity >= 0.60 when capacity evidence is available.

These are bootstrap hypotheses. The observer never creates Cohort 2. Expansion remains a human/community decision until the evidence and security gates justify a higher authority tier.

Public status identity

The workflow maintains one observatory issue using workflow-owned label:

ace:cohort-observer

The title [ACE] Bootstrap Cohort Observatory is for humans, not authority. During migration, an old title-matching issue is adopted only when it already contains an ACE_COHORT_STATE block.

One observatory, in every state

Identity is resolved by asking the API for issues carrying the label in any state, oldest issue number wins, and creation is the last resort:

  1. the oldest issue labelled ace:cohort-observer, open or closed;
  2. otherwise the oldest legacy title match that already carries an ACE_COHORT_STATE block, which is then labelled;
  3. otherwise, and only then, a new observatory.

A closed observatory is reopened in the same call that rewrites its body. The observatory is workflow-owned state rather than a task, so closing it is an issue-tidying mistake, not an instruction to start a second series.

Resolving identity over open issues only is what produced #410 as a fork of #109: the observatory was invisible to the lookup, step 3 fired, and the evidence series that README.md and examples/community/ace-activation-gate-current.example.json cite by issue number split across two numbers. Selecting the newest labelled match then kept every later write on the fork while every document still linked the original. Both selection rules are pinned by tests/test_ace_cohort_observer_contract.py.

When more than one labelled observatory is open at once the run still writes to the oldest and emits core.warning naming all of them, because only a human can decide which duplicate to close.

The machine snapshot includes:

Security boundary

This workflow uses pull_request_target metadata with issue-write capability, so the critical boundary is explicit:

Never check out, import, build, install, evaluate, or execute PR-controlled code in this privileged workflow.

Additional guards:

Anti-Goodhart rule

comment != claim
claim != candidate
candidate != merge
merge != verified causal descendant
bootstrap verified label != complete lineage proof
verified descendant != durable retained contributor
owner/bot activity != external community reproduction

Future versions should add reviewer-time accounting, contributor return/retention, matured parent inventories, and validated lineage integration rather than simply adding event counters.

This site is open source. Improve this page.