mama — measuring and managing autonomy

Names — read this first. mama (lowercase, in prose and code) = measuring and managing autonomy — this method and toolkit: the autonomy-governance component that makes delegated decision rights explicit, assessable, and manageable. This repository is mama's canonical home. MAMA (uppercase) = Measuring Automation & Measuring Autonomy — the broader, integrated framework (umbrella organization) that uses mama alongside automation measurement, project/portfolio context, and business outcomes. The two names are not interchangeable, and this repository is not the MAMA umbrella home. See the integration guide.

Status: v2.0.0 — current release; method pilot / under validation (see the validation plan and limitations).

mama answers three questions for each decision domain of a unit — a person, team, service, department, AI agent, or other sociotechnical unit:

1. Design — What decision authority should this unit have, and why?
2. Measure — What authority is granted, how is it exercised, and how capable and controlled is the unit?
3. Manage — Based on evidence, should the grant expand, contract, be reshaped, or be revoked?

Most organizations manage autonomy implicitly — it emerges from trust, tenure, and accidents of structure, and is therefore felt but not measured, expanded but never designed, and revoked arbitrarily. mama makes autonomy an explicit, designed, measurable property of the organization.

The method in one diagram

flowchart LR
    A[Design<br>decision + blast radius] --> B[Grant<br>autonomy contract]
    B --> C[Measure<br>ledger + telemetry]
    C --> D[Manage<br>expand / contract / re-shape]
    D -->|re-grant, next horizon| B

What's here

Who it is for

Grantors and governance leads who delegate decision authority and need it legible; team leads and unit owners who hold it; risk, security, and AI-governance functions that must evidence control over autonomous behavior; and practitioners assessing AI agents as units. Scores are evidence inputs to human governance decisions — they never authorize a deployment by themselves.

Install

No Python needed — standalone binaries for macOS, Linux, and Windows are attached to each GitHub Release: download autonomy-score-<os>-<arch> and ledger-health-<os>-<arch>, mark them executable, run.

With Python 3.9+ (installs the autonomy-score and ledger-health commands):

pipx install measuring-autonomy    # or: pip install measuring-autonomy

(The distribution keeps the name measuring-autonomy; the method is mama.)

From a clone, the zero-install form keeps working with no packaging step: python3 tools/autonomy_score.py ... (the tools/ scripts are shims over src/measuring_autonomy/). All tools run locally and offline; assessment and ledger data leave your environment only if you send them somewhere.

Quick start: score a team

python3 tools/autonomy_score.py 2 4 3 3 2 2 --current 2

Scores are D1 clarity, D2 capability, D3 guardrails, D4 telemetry, D5 transparency, D6 accountability (0–5 each: definitive levels 1–5, and 0 — the Submarine Line — for no autonomy at all). The tool applies the weakest-link rule — min(guardrails, telemetry, accountability) — and outputs the safe grant level and recommended move.

python3 tools/autonomy_score.py --self-test   # verify the scoring rules

Track granted autonomy in a machine-readable Autonomy Ledger (schema 2 YAML, examples/ledger.yaml) and check its health:

python3 tools/ledger_health.py examples/ledger.yaml   # requires PyYAML (pip install pyyaml)

Try it without a terminal

A dependency-free web version of the assessment lives in site/ (and is published to GitHub Pages when enabled — repo Settings → Pages → Source: GitHub Actions). Open site/index.html locally, tick the items with evidence, and print the scorecard.

Roadmap

See docs/roadmap.md — current phase, shipped releases, and what validation work remains.

Contributing

Feedback, corrections, and real-world applications are very welcome — see CONTRIBUTING.md. Use the issue templates to report framework feedback or submit a case study. Security issues: SECURITY.md. Governance and appeals: GOVERNANCE.md.

License

MIT