/darwin

Signal Collection

Defines the signals Darwin collects to determine project lifecycle phase and ecosystem state.

2026-08 framework / model signal additions

When scanning pyproject.toml, package.json, requirements*.txt, and source imports, treat the following as end-of-life drift signals (feed into ET-10; ET-09 is reserved for Official Spec Conformance drift — see official-fitness-criteria.md):

Detected dependency	Status as of 2026-08	Successor
semantic-kernel (any version)	Maintenance mode since Microsoft Agent Framework 1.0 GA (2026-04-03) — bug/security only	agent-framework (.NET / Python)
pyautogen / autogen-agentchat < 0.12	Superseded by AG2 0.12.0 (2026-04-17), v1.0 line in progress	ag2 ≥ 0.12, or Microsoft Agent Framework
OpenAI Assistants API calls	API retirement 2026-08-26 (announced)	OpenAI Responses API / Agents SDK
ChatGPT configuration or instructions selecting gpt-4o	Retired from ChatGPT on 2026-02-13; the API model remains available	A current ChatGPT model; do not flag API literals without an announced API retirement
langgraph < 1.0	Pre-GA; LangGraph 1.0 shipped 2025-10-22 with no breaking changes	langgraph ≥ 1.0

Sources: https://devblogs.microsoft.com/agent-framework/microsoft-agent-framework-version-1-0/, https://github.com/ag2ai/ag2/releases, https://changelog.langchain.com/announcements/langgraph-1-0-is-now-generally-available, https://help.openai.com/en/articles/20001051.

Signal Categories

1. Git Metrics

Collected from git log, git shortlog, and repository structure.

2. File Structure Signals

Collected from directory listing and file analysis.

3. Activity Log Signals

Collected from .agents/PROJECT.md and .agents/*.md journals.

4. Existing Score Signals

Read from other agents' outputs (never recalculated by Darwin).

Lifecycle Phase Signals

GENESIS

signals:
  file_count: < 50
  test_framework: absent
  commit_count: < 20
  contributors: 1
  ci_cd: absent
  documentation: minimal (README only or empty)
  branches: main/master only
  age: < 30 days
weight: 0.15 per matching signal (max 8 signals)

ACTIVE_BUILD

signals:
  commit_velocity: > 5/day (or > 25/week)
  feat_ratio: > 50% of commits are feat/add
  file_creation_rate: > 3 new files/day
  branch_count: > 3 active feature branches
  contributors: growing (new contributors in last 30 days)
  test_additions: present but trailing features
  ci_cd: basic (may be incomplete)
  merge_frequency: > 3/week
weight: 0.125 per matching signal (max 8 signals)

STABILIZATION

signals:
  refactor_ratio: > 30% of commits are refactor/fix
  test_additions: outpacing feature additions
  code_review_activity: PR comments increasing
  lint_config: present and enforced
  documentation: growing (API docs, guides)
  ci_cd: comprehensive (tests, lint, build)
  feat_velocity: decreasing from peak
  dependency_updates: regular maintenance
weight: 0.125 per matching signal (max 8 signals)

PRODUCTION

signals:
  deploy_configs: present (Docker, k8s, terraform, etc.)
  monitoring: configured (Sentry, DataDog, etc.)
  hotfix_branches: pattern observed (hotfix/*, fix/*)
  release_tags: regular cadence (>= monthly)
  ci_cd: complete (deploy stages included)
  security_scanning: present
  environment_configs: staging/production separation
  incident_handling: runbooks or on-call config present
weight: 0.125 per matching signal (max 8 signals)

MAINTENANCE

signals:
  commit_velocity: < 2/week
  fix_ratio: > 60% of commits are fix/patch
  dependency_updates: primary activity
  feature_additions: rare (< 1/month)
  contributor_count: stable or declining
  age: > 365 days
  documentation: stable (minimal changes)
  ci_cd: stable (no pipeline changes)
weight: 0.125 per matching signal (max 8 signals)

SCALING

signals:
  performance_changes: present (caching, optimization, indexing)
  infrastructure_additions: load balancers, CDN, scaling configs
  load_test_configs: present
  database_optimization: query tuning, partitioning, sharding
  monitoring_expansion: new dashboards, alerts, SLOs
  horizontal_scaling: multi-instance configs
  cost_optimization: resource tuning commits
  capacity_planning: infrastructure docs
weight: 0.125 per matching signal (max 8 signals)

SUNSET

signals:
  last_commit_age: > 60 days
  deprecation_markers: @deprecated tags, README notices
  archive_flags: GitHub archived, "unmaintained" badges
  dependency_status: outdated, security vulnerabilities unpatched
  documentation: "end of life" or migration guides
  ci_cd: failing or disabled
  contributor_count: 0 active
  issues_closed_as_wontfix: increasing
weight: 0.125 per matching signal (max 8 signals)

Signal Collection Process

1. GATHER: Run git commands, scan file structure, read activity logs
2. NORMALIZE: Convert raw values to 0.0-1.0 scores per signal
3. MATCH: Compare normalized scores against phase signal profiles
4. SCORE: Calculate weighted sum per phase
5. SELECT: Choose highest-scoring phase (or mixed if top < 0.60)
6. COMPARE: Check against previous detection for transition events
7. REPORT: Output phase, confidence, and signal breakdown

Normalization Examples

Collection Frequency

Performance note: Full signal collection may take 10-30 seconds on large repositories. Git metric collection is the primary bottleneck. Use cached ECOSYSTEM.md when fresh data is not critical.